Merge pull request #4210 from xianyi/develop

merge develop into 0.3.0 for 0.3.24

.cirrus.yml

@@ -0,0 +1,167 @@
+macos_instance:
+  image: ghcr.io/cirruslabs/macos-monterey-xcode:latest
+
+task:
+  name: AppleM1/LLVM
+  compile_script:
+  - brew install llvm
+  - export PATH=/opt/homebrew/opt/llvm/bin:$PATH
+  - export LDFLAGS="-L/opt/homebrew/opt/llvm/lib"
+  - export CPPFLAGS="-I/opt/homebrew/opt/llvm/include"
+  - make TARGET=VORTEX USE_OPENMP=1 CC=clang
+
+task:
+  name: AppleM1/LLVM/ILP64
+  compile_script:
+  - brew install llvm
+  - export PATH=/opt/homebrew/opt/llvm/bin:$PATH
+  - export LDFLAGS="-L/opt/homebrew/opt/llvm/lib"
+  - export CPPFLAGS="-I/opt/homebrew/opt/llvm/include"
+  - make TARGET=VORTEX USE_OPENMP=1 CC=clang INTERFACE64=1
+
+task:
+  name: AppleM1/LLVM/CMAKE
+  compile_script:
+  - brew install llvm
+  - export PATH=/opt/homebrew/opt/llvm/bin:$PATH
+  - export LDFLAGS="-L/opt/homebrew/opt/llvm/lib"
+  - export CPPFLAGS="-I/opt/homebrew/opt/llvm/include"
+  - mkdir build
+  - cd build
+  - cmake -DTARGET=VORTEX -DCMAKE_C_COMPILER=clang -DBUILD_SHARED_LIBS=ON ..
+  - make
+
+task:
+  name: AppleM1/GCC/MAKE/OPENMP
+  compile_script:
+  - brew install gcc@11
+  - export PATH=/opt/homebrew/bin:$PATH
+  - export LDFLAGS="-L/opt/homebrew/lib"
+  - export CPPFLAGS="-I/opt/homebrew/include"
+  - make CC=gcc-11 FC=gfortran-11 USE_OPENMP=1
+  
+macos_instance:
+  image: ghcr.io/cirruslabs/macos-monterey-xcode:latest
+task:
+  name: AppleM1/LLVM x86_64 xbuild
+  compile_script:
+  - #brew install llvm
+  - export #PATH=/opt/homebrew/opt/llvm/bin:$PATH
+  - export #LDFLAGS="-L/opt/homebrew/opt/llvm/lib"
+  - export #CPPFLAGS="-I/opt/homebrew/opt/llvm/include" 
+  - export ARCHS="i386 x86_64"
+  - export ARCHS_STANDARD="i386 x86_64"
+  - export ARCHS_STANDARD_32_64_BIT="i386 x86_64"
+  - export ARCHS_STANDARD_64_BIT=x86_64
+  - export ARCHS_STANDARD_INCLUDING_64_BIT="i386 x86_64"
+  - export ARCHS_UNIVERSAL_IPHONE_OS="i386 x86_64"
+  - export VALID_ARCHS="i386 x86_64"
+  - xcrun --sdk macosx --show-sdk-path
+  - xcodebuild -version
+  - export CC=/Applications/Xcode-14.0.0.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang
+  - export CFLAGS="-O2 -unwindlib=none -Wno-macro-redefined -isysroot /Applications/Xcode-14.0.0.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX12.3.sdk -arch x86_64"
+  - make TARGET=CORE2 DYNAMIC_ARCH=1 NUM_THREADS=32 HOSTCC=clang NOFORTRAN=1 RANLIB="ls -l"
+  always:
+    config_artifacts:
+      path: "*conf*"
+      type: text/plain
+#  lib_artifacts:
+#    path: "libopenblas*"
+#    type: application/octet-streamm
+
+macos_instance:
+  image: ghcr.io/cirruslabs/macos-monterey-xcode:latest
+task:
+  name: AppleM1/LLVM armv8-ios xbuild
+  compile_script:
+  - #brew install llvm
+  - export #PATH=/opt/homebrew/opt/llvm/bin:$PATH
+  - export #LDFLAGS="-L/opt/homebrew/opt/llvm/lib"
+  - export #CPPFLAGS="-I/opt/homebrew/opt/llvm/include" 
+  - export CC=/Applications/Xcode-14.0.0.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang
+  - export CFLAGS="-O2 -unwindlib=none -Wno-macro-redefined -isysroot /Applications/Xcode-14.0.0.app/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS16.0.sdk -arch arm64 -miphoneos-version-min=10.0"
+  - make TARGET=ARMV8 NUM_THREADS=32 HOSTCC=clang NOFORTRAN=1 CROSS=1
+  always:
+    config_artifacts:
+      path: "*conf*"
+      type: text/plain
+
+macos_instance:
+  image: ghcr.io/cirruslabs/macos-monterey-xcode:latest
+task:
+  name: AppleM1/LLVM armv7-androidndk xbuild
+  compile_script:
+  - #brew install android-ndk
+  - export #PATH=/opt/homebrew/opt/llvm/bin:$PATH
+  - export #LDFLAGS="-L/opt/homebrew/opt/llvm/lib"
+  - export #CPPFLAGS="-I/opt/homebrew/opt/llvm/include" 
+  - find /System/Volumes/Data/opt/homebrew/Caskroom/android-ndk/25b -name "armv7a-linux-androideabi*-ranlib"
+  - #export CC=/Applications/Xcode-13.4.1.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang
+  - #export CFLAGS="-O2 -unwindlib=none -Wno-macro-redefined -isysroot /Applications/Xcode-13.4.1.app/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS16.0.sdk -arch arm64 -miphoneos-version-min=10.0"
+  - export CC=/System/Volumes/Data/opt/homebrew/Caskroom/android-ndk/25b/AndroidNDK8937393.app/Contents/NDK/toolchains/llvm/prebuilt/darwin-x86_64/bin/armv7a-linux-androideabi23-clang
+  - make TARGET=ARMV7 ARM_SOFTFP_ABI=1 NUM_THREADS=32 HOSTCC=clang NOFORTRAN=1 RANLIB="ls -l"
+  always:
+    config_artifacts:
+      path: "*conf*"
+      type: text/plain
+
+task:
+  name: NeoverseN1
+  arm_container:
+    image: node:latest
+  compile_script:
+  - make
+
+  task:
+  name: NeoverseN1-ILP64
+  arm_container:
+    image: node:latest
+  compile_script:
+  - make INTERFACE64=1
+
+task:
+  name: NeoverseN1-OMP
+  arm_container:
+    image: node:latest
+    cpu: 8
+  compile_script:
+  - make USE_OPENMP=1
+
+FreeBSD_task:
+  name: FreeBSD-gcc12
+  freebsd_instance:
+    image_family: freebsd-13-2
+  install_script:
+  - pkg update -f && pkg upgrade -y && pkg install -y gmake gcc
+  compile_script:
+  - ls -l /usr/local/lib
+  - gmake CC=gcc 
+
+
+FreeBSD_task:
+  name: freebsd-gcc12-ilp64
+  freebsd_instance:
+    image_family: freebsd-13-2
+  install_script:
+  - pkg update -f && pkg upgrade -y && pkg install -y gmake gcc
+  compile_script:
+  - ls -l /usr/local/lib
+  - gmake CC=gcc INTERFACE64=1
+
+#task:
+#  name: Windows/LLVM16     --- too slow ---
+#  windows_container:
+#    image: cirrusci/windowsservercore:cmake-2021.12.07
+#  install_script: 
+#    - choco list --localonly
+#    - choco install -y llvm
+#    - # choco install -y cmake --installargs '"ADD_CMAKE_TO_PATH=System"'
+#    - choco install -y ninja
+#    - refreshenv
+#    - cd "c:/Program Files (x86)/Microsoft Visual Studio/2019/BuildTools/VC/Auxiliary/Build"
+#    - vcvarsall x64
+#    - cd "C:\Users\ContainerAdministrator\AppData\Local\Temp\cirrus-ci-build"
+#    - cmake -S . -B build -G "Ninja" -DCMAKE_CXX_COMPILER=clang-cl -DCMAKE_C_COMPILER=clang-cl -DCMAKE_MT=mt -DCMAKE_BUILD_TYPE=Release
+#    - cd build
+#    - cmake --build .
+#    - ctest

.github/workflows/c910v.yml

@@ -0,0 +1,121 @@
+name: c910v qemu test
+
+on: [push, pull_request]
+
+permissions:
+  contents: read # to fetch code (actions/checkout)
+
+jobs:
+  TEST:
+    runs-on: ubuntu-latest
+    env:
+      xuetie_toolchain: https://occ-oss-prod.oss-cn-hangzhou.aliyuncs.com/resource//1663142514282
+      toolchain_file_name: Xuantie-900-gcc-linux-5.10.4-glibc-x86_64-V2.6.1-20220906.tar.gz
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - target: RISCV64_GENERIC
+            triple: riscv64-linux-gnu
+            apt_triple: riscv64-linux-gnu
+            opts: NO_SHARED=1 TARGET=RISCV64_GENERIC
+          - target: C910V
+            triple: riscv64-unknown-linux-gnu
+            apt_triple: riscv64-linux-gnu
+            opts: NO_SHARED=1 TARGET=C910V
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v3
+
+      - name: install build deps
+        run: |
+          sudo apt-get update
+          sudo apt-get install autoconf automake autotools-dev ninja-build make ccache \
+          gcc-${{ matrix.apt_triple }} gfortran-${{ matrix.apt_triple }} libgomp1-riscv64-cross
+
+      - name: checkout qemu
+        uses: actions/checkout@v3
+        with:
+          repository: T-head-Semi/qemu
+          path: qemu
+          ref: 1e692ebb43d396c52352406323fc782c1ac99a42
+
+      - name: build qemu
+        run: |
+          # Force use c910v qemu-user
+          wget https://github.com/revyos/qemu/commit/5164bca5a4bcde4534dc1a9aa3a7f619719874cf.patch
+          cd qemu
+          patch -p1 < ../5164bca5a4bcde4534dc1a9aa3a7f619719874cf.patch
+          ./configure --prefix=$GITHUB_WORKSPACE/qemu-install --target-list=riscv64-linux-user --disable-system
+          make -j$(nproc)
+          make install
+
+      - name: Compilation cache
+        uses: actions/cache@v3
+        with:
+          path: ~/.ccache
+          key: ccache-${{ runner.os }}-${{ matrix.target }}-${{ github.ref }}-${{ github.sha }}
+          restore-keys: |
+            ccache-${{ runner.os }}-${{ matrix.target }}-${{ github.ref }}
+            ccache-${{ runner.os }}-${{ matrix.target }}
+
+      - name: Configure ccache
+        run: |
+          test -d ~/.ccache || mkdir -p ~/.ccache
+          echo "max_size = 300M" > ~/.ccache/ccache.conf
+          echo "compression = true" >> ~/.ccache/ccache.conf
+          ccache -s
+
+      - name: build OpenBLAS
+        run: |
+            wget ${xuetie_toolchain}/${toolchain_file_name}
+            tar -xvf ${toolchain_file_name} -C /opt
+            export PATH="/opt/Xuantie-900-gcc-linux-5.10.4-glibc-x86_64-V2.6.1/bin:$PATH"
+
+            make CC='ccache ${{ matrix.triple }}-gcc -static' FC='ccache ${{ matrix.triple }}-gfortran -static' ${{ matrix.opts }} HOSTCC='ccache gcc' -j$(nproc)
+
+      - name: test
+        run: |
+          export PATH=$GITHUB_WORKSPACE/qemu-install/bin/:$PATH
+          qemu-riscv64 ./utest/openblas_utest
+          OPENBLAS_NUM_THREADS=2 qemu-riscv64 ./ctest/xscblat1
+          OPENBLAS_NUM_THREADS=2 qemu-riscv64 ./ctest/xdcblat1
+          OPENBLAS_NUM_THREADS=2 qemu-riscv64 ./ctest/xccblat1
+          OPENBLAS_NUM_THREADS=2 qemu-riscv64 ./ctest/xzcblat1
+          OPENBLAS_NUM_THREADS=2 qemu-riscv64 ./ctest/xscblat2 < ./ctest/sin2
+          OPENBLAS_NUM_THREADS=2 qemu-riscv64 ./ctest/xdcblat2 < ./ctest/din2
+          OPENBLAS_NUM_THREADS=2 qemu-riscv64 ./ctest/xccblat2 < ./ctest/cin2
+          OPENBLAS_NUM_THREADS=2 qemu-riscv64 ./ctest/xzcblat2 < ./ctest/zin2
+          OPENBLAS_NUM_THREADS=2 qemu-riscv64 ./ctest/xscblat3 < ./ctest/sin3
+          OPENBLAS_NUM_THREADS=2 qemu-riscv64 ./ctest/xdcblat3 < ./ctest/din3
+          OPENBLAS_NUM_THREADS=2 qemu-riscv64 ./ctest/xccblat3 < ./ctest/cin3
+          OPENBLAS_NUM_THREADS=2 qemu-riscv64 ./ctest/xzcblat3 < ./ctest/zin3
+          OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 qemu-riscv64 ./test/sblat1
+          OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 qemu-riscv64 ./test/dblat1
+          OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 qemu-riscv64 ./test/cblat1
+          OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 qemu-riscv64 ./test/zblat1
+          OPENBLAS_NUM_THREADS=2 qemu-riscv64 ./test/sblat1
+          OPENBLAS_NUM_THREADS=2 qemu-riscv64 ./test/dblat1
+          OPENBLAS_NUM_THREADS=2 qemu-riscv64 ./test/cblat1
+          OPENBLAS_NUM_THREADS=2 qemu-riscv64 ./test/zblat1
+          rm -f ./test/?BLAT2.SUMM
+          OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 qemu-riscv64 ./test/sblat2 < ./test/sblat2.dat
+          OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 qemu-riscv64 ./test/dblat2 < ./test/dblat2.dat
+          OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 qemu-riscv64 ./test/cblat2 < ./test/cblat2.dat
+          OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 qemu-riscv64 ./test/zblat2 < ./test/zblat2.dat
+          rm -f ./test/?BLAT2.SUMM
+          OPENBLAS_NUM_THREADS=2 qemu-riscv64 ./test/sblat2 < ./test/sblat2.dat
+          OPENBLAS_NUM_THREADS=2 qemu-riscv64 ./test/dblat2 < ./test/dblat2.dat
+          OPENBLAS_NUM_THREADS=2 qemu-riscv64 ./test/cblat2 < ./test/cblat2.dat
+          OPENBLAS_NUM_THREADS=2 qemu-riscv64 ./test/zblat2 < ./test/zblat2.dat
+          rm -f ./test/?BLAT3.SUMM
+          OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 qemu-riscv64 ./test/sblat3 < ./test/sblat3.dat
+          OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 qemu-riscv64 ./test/dblat3 < ./test/dblat3.dat
+          OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 qemu-riscv64 ./test/cblat3 < ./test/cblat3.dat
+          OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 qemu-riscv64 ./test/zblat3 < ./test/zblat3.dat
+          rm -f ./test/?BLAT3.SUMM
+          OPENBLAS_NUM_THREADS=2 qemu-riscv64 ./test/sblat3 < ./test/sblat3.dat
+          OPENBLAS_NUM_THREADS=2 qemu-riscv64 ./test/dblat3 < ./test/dblat3.dat
+          OPENBLAS_NUM_THREADS=2 qemu-riscv64 ./test/cblat3 < ./test/cblat3.dat
+          OPENBLAS_NUM_THREADS=2 qemu-riscv64 ./test/zblat3 < ./test/zblat3.dat

.github/workflows/dynamic_arch.yml

@@ -151,40 +151,53 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        msystem: [MINGW64, MINGW32, CLANG64]
+        msystem: [MINGW64, MINGW32, CLANG64, CLANG32]
         idx: [int32, int64]
         build-type: [Release]
         include:
           - msystem: MINGW64
             idx: int32
             target-prefix: mingw-w64-x86_64
-            fc-pkg: mingw-w64-x86_64-gcc-fortran
+            fc-pkg: fc
           - msystem: MINGW32
             idx: int32
             target-prefix: mingw-w64-i686
-            fc-pkg: mingw-w64-i686-gcc-fortran
+            fc-pkg: fc
           - msystem: CLANG64
             idx: int32
             target-prefix: mingw-w64-clang-x86_64
+            fc-pkg: fc
+            # Compiling with Flang 16 seems to cause test errors on machines
+            # with AVX512 instructions. Revisit after MSYS2 distributes Flang 17.
+            no-avx512-flags: -DNO_AVX512=1
+          - msystem: CLANG32
+            idx: int32
+            target-prefix: mingw-w64-clang-i686
+            fc-pkg: cc
             c-lapack-flags: -DC_LAPACK=ON
           - msystem: MINGW64
             idx: int64
             idx64-flags: -DBINARY=64 -DINTERFACE64=1
             target-prefix: mingw-w64-x86_64
-            fc-pkg: mingw-w64-x86_64-gcc-fortran
+            fc-pkg: fc
           - msystem: CLANG64
             idx: int64
             idx64-flags: -DBINARY=64 -DINTERFACE64=1
             target-prefix: mingw-w64-clang-x86_64
-            c-lapack-flags: -DC_LAPACK=ON
+            fc-pkg: fc
+            # Compiling with Flang 16 seems to cause test errors on machines
+            # with AVX512 instructions. Revisit after MSYS2 distributes Flang 17.
+            no-avx512-flags: -DNO_AVX512=1
           - msystem: MINGW64
             idx: int32
             target-prefix: mingw-w64-x86_64
-            fc-pkg: mingw-w64-x86_64-gcc-fortran
+            fc-pkg: fc
             build-type: None
         exclude:
           - msystem: MINGW32
             idx: int64
+          - msystem: CLANG32
+            idx: int64
 
     defaults:
       run:
@@ -209,7 +222,7 @@ jobs:
           install: >-
             base-devel
             ${{ matrix.target-prefix }}-cc
-            ${{ matrix.fc-pkg }}
+            ${{ matrix.target-prefix }}-${{ matrix.fc-pkg }}
             ${{ matrix.target-prefix }}-cmake
             ${{ matrix.target-prefix }}-ninja
             ${{ matrix.target-prefix }}-ccache
@@ -217,14 +230,21 @@ jobs:
       - name: Checkout repository
         uses: actions/checkout@v3
 
-      - name: Compilation cache
-        uses: actions/cache@v3
-        with:
-          # It looks like this path needs to be hard-coded.
-          path: C:/msys64/home/runneradmin/.ccache
+      - name: Prepare ccache
+        # Get cache location of ccache
+        # Create key that is used in action/cache/restore and action/cache/save steps
+        id: ccache-prepare
+        run: |
+          echo "ccachedir=$(cygpath -m $(ccache -k cache_dir))" >> $GITHUB_OUTPUT
           # We include the commit sha in the cache key, as new cache entries are
           # only created if there is no existing entry for the key yet.
-          key: ccache-msys2-${{ matrix.msystem }}-${{ matrix.idx }}-${{ matrix.build-type }}-${{ github.ref }}-${{ github.sha }}
+          echo "key=ccache-msys2-${{ matrix.msystem }}-${{ matrix.idx }}-${{ matrix.build-type }}-${{ github.ref }}-${{ github.sha }}" >> $GITHUB_OUTPUT
+
+      - name: Restore ccache
+        uses: actions/cache/restore@v3
+        with:
+          path: ${{ steps.ccache-prepare.outputs.ccachedir }}
+          key: ${{ steps.ccache-prepare.outputs.key }}
           # Restore a matching ccache cache entry. Prefer same branch.
           restore-keys: |
             ccache-msys2-${{ matrix.msystem }}-${{ matrix.idx }}-${{ matrix.build-type }}-${{ github.ref }}
@@ -234,9 +254,10 @@ jobs:
         # Limit the maximum size and switch on compression to avoid exceeding the total disk or cache quota.
         run: |
           which ccache
-          test -d ~/.ccache || mkdir -p ~/.ccache
-          echo "max_size = 250M" > ~/.ccache/ccache.conf
-          echo "compression = true" >> ~/.ccache/ccache.conf
+          test -d ${{ steps.ccache-prepare.outputs.ccachedir }} || mkdir -p ${{ steps.ccache-prepare.outputs.ccachedir }}
+          echo "max_size = 250M" > ${{ steps.ccache-prepare.outputs.ccachedir }}/ccache.conf
+          echo "compression = true" >> ${{ steps.ccache-prepare.outputs.ccachedir }}/ccache.conf
+          ccache -p
           ccache -s
           echo $HOME
           cygpath -w $HOME
@@ -253,6 +274,7 @@ jobs:
                 -DTARGET=CORE2 \
                 ${{ matrix.idx64-flags }} \
                 ${{ matrix.c-lapack-flags }} \
+                ${{ matrix.no-avx512-flags }} \
                 -DCMAKE_C_COMPILER_LAUNCHER=ccache \
                 -DCMAKE_Fortran_COMPILER_LAUNCHER=ccache \
                 ..
@@ -264,10 +286,30 @@ jobs:
         continue-on-error: true
         run: ccache -s
 
+      - name: Save ccache
+        # Save the cache after we are done (successfully) building
+        uses: actions/cache/save@v3
+        with:
+          path: ${{ steps.ccache-prepare.outputs.ccachedir }}
+          key: ${{ steps.ccache-prepare.outputs.key }}
+
       - name: Run tests
+        id: run-ctest
         timeout-minutes: 60
         run: cd build && ctest
 
+      - name: Re-run tests
+        if: always() && (steps.run-ctest.outcome == 'failure')
+        timeout-minutes: 60
+        run: |
+          cd build
+          echo "::group::Re-run ctest"
+          ctest --rerun-failed --output-on-failure || true
+          echo "::endgroup::"
+          echo "::group::Log from these tests"
+          [ ! -f Testing/Temporary/LastTest.log ] || cat Testing/Temporary/LastTest.log
+          echo "::endgroup::"
+
 
   cross_build:
     runs-on: ubuntu-22.04
@@ -295,6 +337,7 @@ jobs:
 
       - name: Install Dependencies
         run: |
+          sudo apt-get update
           sudo apt-get install -y ccache gcc-${{ matrix.triple }} gfortran-${{ matrix.triple }} libgomp1-${{ matrix.target }}-cross
 
       - name: Compilation cache

.github/workflows/loongarch64.yml

@@ -0,0 +1,110 @@
+name: loongarch64 qemu test
+
+on: [push, pull_request]
+
+jobs:
+  TEST:
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - target: LOONGSONGENERIC
+            triple:  loongarch64-unknown-linux-gnu
+            opts: NO_SHARED=1 TARGET=LOONGSONGENERIC
+          - target: LOONGSON3R5
+            triple: loongarch64-unknown-linux-gnu
+            opts: NO_SHARED=1 TARGET=LOONGSON3R5
+          - target: LOONGSON2K1000
+            triple: loongarch64-unknown-linux-gnu
+            opts: NO_SHARED=1 TARGET=LOONGSON2K1000
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v3
+
+      - name: Install APT deps
+        run: |
+          sudo add-apt-repository ppa:savoury1/virtualisation
+          sudo apt-get update
+          sudo apt-get install autoconf automake autotools-dev ninja-build make ccache \
+          qemu-user-static
+
+      - name: Download and install loongarch64-toolchain
+        run: |
+          wget https://github.com/loongson/build-tools/releases/download/2022.09.06/loongarch64-clfs-7.3-cross-tools-gcc-glibc.tar.xz
+          tar -xf loongarch64-clfs-7.3-cross-tools-gcc-glibc.tar.xz -C /opt
+
+      - name: Set env
+        run: |
+          echo "LD_LIBRARY_PATH=/opt/cross-tools/target/usr/lib64:/opt/cross-tools/loongarch64-unknown-linux-gnu/lib64:$LD_LIBRARY_PATH" >> $GITHUB_ENV
+          echo "PATH=$GITHUB_WORKSPACE:/opt/cross-tools/bin:$PATH" >> $GITHUB_ENV
+
+      - name: Compilation cache
+        uses: actions/cache@v3
+        with:
+          path: ~/.ccache
+          key: ccache-${{ runner.os }}-${{ matrix.target }}-${{ github.ref }}-${{ github.sha }}
+          restore-keys: |
+            ccache-${{ runner.os }}-${{ matrix.target }}-${{ github.ref }}
+            ccache-${{ runner.os }}-${{ matrix.target }}
+
+      - name: Configure ccache
+        run: |
+          test -d ~/.ccache || mkdir -p ~/.ccache
+          echo "max_size = 300M" > ~/.ccache/ccache.conf
+          echo "compression = true" >> ~/.ccache/ccache.conf
+          ccache -s
+
+      - name: Disable utest dsdot:dsdot_n_1
+        run: |
+          echo -n > utest/test_dsdot.c
+          echo "Due to the qemu versions 7.2 causing utest cases to fail,"
+          echo "the utest dsdot:dsdot_n_1 have been temporarily disabled."
+
+      - name: Build OpenBLAS
+        run: make CC='ccache ${{ matrix.triple }}-gcc -static' FC='ccache ${{ matrix.triple }}-gfortran -static' ${{ matrix.opts }} HOSTCC='ccache gcc' -j$(nproc)
+
+      - name: Test
+        run: |
+          qemu-loongarch64-static ./utest/openblas_utest
+          OPENBLAS_NUM_THREADS=2 qemu-loongarch64-static ./ctest/xscblat1
+          OPENBLAS_NUM_THREADS=2 qemu-loongarch64-static ./ctest/xdcblat1
+          OPENBLAS_NUM_THREADS=2 qemu-loongarch64-static ./ctest/xccblat1
+          OPENBLAS_NUM_THREADS=2 qemu-loongarch64-static ./ctest/xzcblat1
+          OPENBLAS_NUM_THREADS=2 qemu-loongarch64-static ./ctest/xscblat2 < ./ctest/sin2
+          OPENBLAS_NUM_THREADS=2 qemu-loongarch64-static ./ctest/xdcblat2 < ./ctest/din2
+          OPENBLAS_NUM_THREADS=2 qemu-loongarch64-static ./ctest/xccblat2 < ./ctest/cin2
+          OPENBLAS_NUM_THREADS=2 qemu-loongarch64-static ./ctest/xzcblat2 < ./ctest/zin2
+          OPENBLAS_NUM_THREADS=2 qemu-loongarch64-static ./ctest/xscblat3 < ./ctest/sin3
+          OPENBLAS_NUM_THREADS=2 qemu-loongarch64-static ./ctest/xdcblat3 < ./ctest/din3
+          OPENBLAS_NUM_THREADS=2 qemu-loongarch64-static ./ctest/xccblat3 < ./ctest/cin3
+          OPENBLAS_NUM_THREADS=2 qemu-loongarch64-static ./ctest/xzcblat3 < ./ctest/zin3
+          OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 qemu-loongarch64-static ./test/sblat1
+          OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 qemu-loongarch64-static ./test/dblat1
+          OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 qemu-loongarch64-static ./test/cblat1
+          OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 qemu-loongarch64-static ./test/zblat1
+          OPENBLAS_NUM_THREADS=2 qemu-loongarch64-static ./test/sblat1
+          OPENBLAS_NUM_THREADS=2 qemu-loongarch64-static ./test/dblat1
+          OPENBLAS_NUM_THREADS=2 qemu-loongarch64-static ./test/cblat1
+          OPENBLAS_NUM_THREADS=2 qemu-loongarch64-static ./test/zblat1
+          rm -f ./test/?BLAT2.SUMM
+          OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 qemu-loongarch64-static ./test/sblat2 < ./test/sblat2.dat
+          OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 qemu-loongarch64-static ./test/dblat2 < ./test/dblat2.dat
+          OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 qemu-loongarch64-static ./test/cblat2 < ./test/cblat2.dat
+          OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 qemu-loongarch64-static ./test/zblat2 < ./test/zblat2.dat
+          rm -f ./test/?BLAT2.SUMM
+          OPENBLAS_NUM_THREADS=2 qemu-loongarch64-static ./test/sblat2 < ./test/sblat2.dat
+          OPENBLAS_NUM_THREADS=2 qemu-loongarch64-static ./test/dblat2 < ./test/dblat2.dat
+          OPENBLAS_NUM_THREADS=2 qemu-loongarch64-static ./test/cblat2 < ./test/cblat2.dat
+          OPENBLAS_NUM_THREADS=2 qemu-loongarch64-static ./test/zblat2 < ./test/zblat2.dat
+          rm -f ./test/?BLAT3.SUMM
+          OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 qemu-loongarch64-static ./test/sblat3 < ./test/sblat3.dat
+          OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 qemu-loongarch64-static ./test/dblat3 < ./test/dblat3.dat
+          OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 qemu-loongarch64-static ./test/cblat3 < ./test/cblat3.dat
+          OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 qemu-loongarch64-static ./test/zblat3 < ./test/zblat3.dat
+          rm -f ./test/?BLAT3.SUMM
+          OPENBLAS_NUM_THREADS=2 qemu-loongarch64-static ./test/sblat3 < ./test/sblat3.dat
+          OPENBLAS_NUM_THREADS=2 qemu-loongarch64-static ./test/dblat3 < ./test/dblat3.dat
+          OPENBLAS_NUM_THREADS=2 qemu-loongarch64-static ./test/cblat3 < ./test/cblat3.dat
+          OPENBLAS_NUM_THREADS=2 qemu-loongarch64-static ./test/zblat3 < ./test/zblat3.dat

.gitignore

@@ -14,6 +14,7 @@ lapack-3.4.2
 lapack-3.4.2.tgz
 lapack-netlib/make.inc
 lapack-netlib/lapacke/include/lapacke_mangling.h
+lapack-netlib/SRC/la_constants.mod
 lapack-netlib/TESTING/testing_results.txt
 lapack-netlib/INSTALL/test*
 lapack-netlib/TESTING/xeigtstc
@@ -71,6 +72,7 @@ test/SBLAT3.SUMM
 test/ZBLAT2.SUMM
 test/ZBLAT3.SUMM
 test/SHBLAT3.SUMM
+test/SBBLAT3.SUMM
 test/cblat1
 test/cblat2
 test/cblat3
@@ -81,6 +83,7 @@ test/sblat1
 test/sblat2
 test/sblat3
 test/test_shgemm
+test/test_sbgemm
 test/zblat1
 test/zblat2
 test/zblat3

CMakeLists.txt

@@ -8,7 +8,7 @@ project(OpenBLAS C ASM)
 
 set(OpenBLAS_MAJOR_VERSION 0)
 set(OpenBLAS_MINOR_VERSION 3)
-set(OpenBLAS_PATCH_VERSION 23)
+set(OpenBLAS_PATCH_VERSION 23.dev)
 
 set(OpenBLAS_VERSION "${OpenBLAS_MAJOR_VERSION}.${OpenBLAS_MINOR_VERSION}.${OpenBLAS_PATCH_VERSION}")
 
@@ -20,6 +20,8 @@ include(CMakePackageConfigHelpers)
 #######
 option(BUILD_WITHOUT_LAPACK "Do not build LAPACK and LAPACKE (Only BLAS or CBLAS)" OFF)
 
+option(BUILD_LAPACK_DEPRECATED "When building LAPACK, include also some older, deprecated routines" ON)
+
 option(BUILD_TESTING "Build LAPACK testsuite when building LAPACK" ON)
 
 option(C_LAPACK "Build LAPACK from C sources instead of the original Fortran" OFF)
@@ -309,19 +311,25 @@ endif()
 
 #if (MSVC OR NOT NOFORTRAN)
 if (NOT NO_CBLAS)
+  if (NOT ONLY_CBLAS)
   # Broken without fortran on unix
-  add_subdirectory(utest)
+    add_subdirectory(utest)
+endif()
 endif()
 
 if (NOT NOFORTRAN)
+ if (NOT ONLY_CBLAS)
   # Build test and ctest
   add_subdirectory(test)
+ endif()
   if (BUILD_TESTING)
     add_subdirectory(lapack-netlib/TESTING)
   endif()
 endif()
   if(NOT NO_CBLAS)
+   if (NOT ONLY_CBLAS)
     add_subdirectory(ctest)
+   endif()
   endif()
   if (CPP_THREAD_SAFETY_TEST OR CPP_THREAD_SAFETY_GEMV)
     add_subdirectory(cpp_thread_test)
@@ -398,15 +406,45 @@ if (BUILD_SHARED_LIBS AND NOT ${SYMBOLPREFIX}${SYMBOLSUFFIX} STREQUAL "")
     message(STATUS "adding suffix ${SYMBOLSUFFIX} to names of exported symbols in ${OpenBLAS_LIBNAME}")
   endif()
 
+  if (${BUILD_LAPACK_DEPRECATED})
+	  set (BLD 1)
+  else ()
+	  set (BLD 0)
+  endif()
+  if (${BUILD_BFLOAT16})
+	  set (BBF16 1)
+  else ()
+	  set (BBF16 0)
+  endif()
+  if (${BUILD_SINGLE})
+	  set (BS 1)
+  else ()
+	  set (BS 0)
+  endif()
+  if (${BUILD_DOUBLE})
+	  set (BD 1)
+  else ()
+	  set (BD 0)
+  endif()
+  if (${BUILD_COMPLEX})
+	  set (BC 1)
+  else ()
+	  set (BC 0)
+  endif()
+  if (${BUILD_COMPLEX16})
+	  set (BZ 1)
+  else ()
+	  set (BZ 0)
+  endif()
   if (NOT USE_PERL)
   add_custom_command(TARGET ${OpenBLAS_LIBNAME}_shared POST_BUILD
-    COMMAND  ${PROJECT_SOURCE_DIR}/exports/gensymbol "objcopy" "${ARCH}" "${BU}" "${EXPRECISION_IN}" "${NO_CBLAS_IN}" "${NO_LAPACK_IN}" "${NO_LAPACKE_IN}" "${NEED2UNDERSCORES_IN}" "${ONLY_CBLAS_IN}" \"${SYMBOLPREFIX}\" \"${SYMBOLSUFFIX}\" "${BUILD_LAPACK_DEPRECATED}" "${BUILD_BFLOAT16}" "${BUILD_SINGLE}" "${BUILD_DOUBLE}" "${BUILD_COMPLEX}" "${BUILD_COMPLEX16}" > ${PROJECT_BINARY_DIR}/objcopy.def
+	  COMMAND  ${PROJECT_SOURCE_DIR}/exports/gensymbol "objcopy" "${ARCH}" "${BU}" "${EXPRECISION_IN}" "${NO_CBLAS_IN}" "${NO_LAPACK_IN}" "${NO_LAPACKE_IN}" "${NEED2UNDERSCORES_IN}" "${ONLY_CBLAS_IN}" \"${SYMBOLPREFIX}\" \"${SYMBOLSUFFIX}\" "${BLD}" "${BBF16}" "${BS}" "${BD}" "${BC}" "${BZ}" > ${PROJECT_BINARY_DIR}/objcopy.def
     COMMAND objcopy -v --redefine-syms ${PROJECT_BINARY_DIR}/objcopy.def  ${PROJECT_BINARY_DIR}/lib/lib${OpenBLAS_LIBNAME}.so
     COMMENT "renaming symbols"
     )
   else()
   add_custom_command(TARGET ${OpenBLAS_LIBNAME}_shared POST_BUILD
-    COMMAND perl ${PROJECT_SOURCE_DIR}/exports/gensymbol.pl "objcopy" "${ARCH}" "${BU}" "${EXPRECISION_IN}" "${NO_CBLAS_IN}" "${NO_LAPACK_IN}" "${NO_LAPACKE_IN}" "${NEED2UNDERSCORES_IN}" "${ONLY_CBLAS_IN}" \"${SYMBOLPREFIX}\" \"${SYMBOLSUFFIX}\" "${BUILD_LAPACK_DEPRECATED}" "${BUILD_BFLOAT16}" "${BUILD_SINGLE}" "${BUILD_DOUBLE}" "${BUILD_COMPLEX}" "${BUILD_COMPLEX16}" > ${PROJECT_BINARY_DIR}/objcopy.def
+    COMMAND perl ${PROJECT_SOURCE_DIR}/exports/gensymbol.pl "objcopy" "${ARCH}" "${BU}" "${EXPRECISION_IN}" "${NO_CBLAS_IN}" "${NO_LAPACK_IN}" "${NO_LAPACKE_IN}" "${NEED2UNDERSCORES_IN}" "${ONLY_CBLAS_IN}" \"${SYMBOLPREFIX}\" \"${SYMBOLSUFFIX}\" "${BLD}" "${BBF16}" "${BS}" "${BD}" "${BC}" "${BZ}" > ${PROJECT_BINARY_DIR}/objcopy.def
     COMMAND objcopy -v --redefine-syms ${PROJECT_BINARY_DIR}/objcopy.def  ${PROJECT_BINARY_DIR}/lib/lib${OpenBLAS_LIBNAME}.so
     COMMENT "renaming symbols"
     )
@@ -511,9 +549,8 @@ configure_file(${PROJECT_SOURCE_DIR}/cmake/openblas.pc.in ${PROJECT_BINARY_DIR}/
 install (FILES ${PROJECT_BINARY_DIR}/openblas${SUFFIX64}.pc DESTINATION ${CMAKE_INSTALL_LIBDIR}/pkgconfig/)
 
 
-# GNUInstallDirs "DATADIR" wrong here; CMake search path wants "share".
 set(PN OpenBLAS)
-set(CMAKECONFIG_INSTALL_DIR "share/cmake/${PN}${SUFFIX64}")
+set(CMAKECONFIG_INSTALL_DIR "${CMAKE_INSTALL_LIBDIR}/cmake/${PN}${SUFFIX64}")
 configure_package_config_file(cmake/${PN}Config.cmake.in
                               "${CMAKE_CURRENT_BINARY_DIR}/${PN}${SUFFIX64}Config.cmake"
                               INSTALL_DESTINATION ${CMAKECONFIG_INSTALL_DIR})

CONTRIBUTORS.md

@@ -23,6 +23,9 @@
   * Optimization on AMD Piledriver
   * Optimization on Intel Haswell
 
+* Chris Sidebottom <chris.sidebottom@arm.com>
+  * Optimizations and other improvements targeting AArch64
+
 ## Previous Developers
 
 * Zaheer Chothia <zaheer.chothia@gmail.com>
@@ -212,4 +215,4 @@ In chronological order:
   * [2022-03] Support RISC-V Vector Intrinisc 1.0 version.
   
 * Pablo Romero <https://github.com/pablorcum>
-  * [2022-08] Fix building from sources for QNX
+  * [2022-08] Fix building from sources for QNX

Changelog.txt

@@ -1,4 +1,104 @@
 OpenBLAS ChangeLog
+====================================================================
+Version 0.3.24
+ 03-Sep-2023
+
+general:
+   - declared the arguments of cblas_xerbla as const (in accordance with the reference implementation 
+     and others, the previous discrepancy appears to have dated back to GotoBLAS)
+   - fixed the implementation of ?GEMMT that was added in 0.3.23
+   - made cpu-specific SWITCH_RATIO parameters for GEMM available to DYNAMIC_ARCH builds
+   - fixed application of SYMBOLSUFFIX in CMAKE builds
+   - fixed missing SSYCONVF function in the shared library
+   - fixed parallel build logic used with gmake
+   - added support for compilation with LLVM17, in particular its new Fortran compiler
+   - added support for CMAKE builds using the NVIDIA HPC compiler
+   - fixed INTERFACE64 builds with CMAKE and the f95 Fortran compiler
+   - fixed cross-build detection and management in c_check
+   - disabled building of the tests with CMAKE when ONLY_CBLAS is defined
+   - fixed several issues with the handling of runtime limits on the number of OPENMP threads
+   - corrected the error code returned by SGEADD/DGEADD when LDA is too small
+   - corrected the error code returned by IMATCOPY when LDB is too small
+   - updated ?NRM2 to support negative increment values (as introduced in release 3.10 
+     of the reference BLAS)
+   - fixed OpenMP builds with CLANG for the case where libomp is not in a standard location
+   - fixed a potential overwrite of unrelated memory during thread initialisation on startup
+   - fixed a potential integer overflow in the multithreading threshold for ?SYMM/?SYRK
+   - fixed build of the LAPACKE interfaces for the LAPACK 3.11.0 ?TRSYL functions added in 0.3.22
+   - fixed installation of .cmake files in concurrent 32 and 64bit builds with CMAKE
+   - applied additions and corrections from the development branch of Reference-LAPACK:
+   - fixed actual arguments passed to a number of LAPACK functions (from Reference-LAPACK PR 885)
+   - fixed workspace query results in LAPACK ?SYTRF/?TRECV3 (from Reference-LAPACK PR 883)
+   - fixed derivation of the UPLO parameter in LAPACKE_?larfb (from Reference-LAPACK PR 878)
+   - fixed a crash in LAPACK ?GELSDD on NRHS=0 (from Reference-LAPACK PR 876)
+   - added new LAPACK utility functions CRSCL and ZRSCL (from Reference-LAPACK PR 839)
+	- corrected the order of eigenvalues for 2x2 matrices in ?STEMR (Reference-LAPACK PR 867)
+	- removed spurious reference to OpenMP variables outside OpenMP contexts (Reference-LAPACK PR 860)
+	- updated file comments on use of LAMBDA variable in LAPACK (Reference-LAPACK PR 852)
+	- fixed documentation of LAPACK SLASD0/DLASD0 (Reference-LAPACK PR 855)
+	- fixed confusing use of "minor" in LAPACK documentation (Reference-LAPACK PR 849)
+	- added new LAPACK functions ?GEDMD for dynamic mode decomposition (Reference-LAPACK PR 736)
+	- fixed potential stack overflows in the EIG part of the LAPACK testsuite (Reference-LAPACK PR 854)
+	- applied small improvements to the variants of Cholesky and QR functions (Reference-LAPACK PR 847)
+	- removed unused variables from LAPACK ?BDSQR (Reference-LAPACK PR 832)
+	- fixed a potential crash on allocation failure in LAPACKE SGEESX/DGEESX (Reference-LAPACK PR 836)
+	- added a quick return from SLARUV/DLARUV for N < 1 (Reference-LAPACK PR 837)
+	- updated function descriptions in LAPACK ?GEGS/?GEGV (Reference-LAPACK PR 831)
+	- improved algorithm description in ?GELSY (Reference-LAPACK PR 833)
+	- fixed scaling in LAPACK STGSNA/DTGSNA (Reference-LAPACK PR 830)
+	- fixed crash in LAPACKE_?geqrt with row-major data (Reference-LAPACK PR 768)
+	- added LAPACKE interfaces for C/ZUNHR_COL and S/DORHR_COL (Reference-LAPACK PR 827)
+	- added error exit tests for SYSV/SYTD2/GEHD2 to the testsuite (Reference-LAPACK PR 795)
+	- fixed typos in LAPACK source and comments (Reference-LAPACK PRs 809,811,812,814,820)
+	- adopt refactored ?GEBAL implementation (Reference-LAPACK PR 808)
+
+x86_64:
+   - added cpu model autodetection for Intel Alder Lake N
+   - added activation of the AMX tile to the Sapphire Rapids SBGEMM kernel
+   - worked around miscompilations of GEMV/SYMV kernels by gcc's tree-vectorizer
+   - fixed compilation of Cooperlake and Sapphire Rapids kernels with CLANG
+   - fixed runtime detection of Cooperlake and Sapphire Rapids in DYNAMIC_ARCH
+   - fixed feature-based cputype fallback in DYNAMIC_ARCH
+   - added support for building the AVX512 kernels with the NVIDIA HPC compiler
+   - corrected ZAXPY result on old pre-AVX hardware for the INCX=0 case
+   - fixed a potential use of uninitialized variables in ZTRSM
+
+ARM64:
+   - added cpu model autodetection for Apple M2
+   - fixed wrong results of CGEMM/CTRMM/DNRM2 under OSX (use of reserved register)
+   - added support for building the SVE kernels with the NVIDIA HPC compiler
+   - added support for building the SVE kernels with the Apple Clang compiler
+   - fixed compiler option handling for building the SVE kernels with LLVM
+   - implemented SWITCH_RATIO parameter for improved GEMM performance on Neoverse
+   - activated SVE SGEMM and DGEMM kernels for Neoverse V1
+   - improved performance of the SVE CGEMM and ZGEMM kernels on Neoverse V1
+   - improved kernel selection for the ARMV8SVE target and added it to DYNAMIC_ARCH
+   - fixed runtime check for SVE availability in DYNAMIC_ARCH builds to take OS or
+     container restrictions into account
+   - fixed a potential use of uninitialized variables in ZTRSM
+   - fix a potential misdetection of ARMV8 hardware as 32bit in CMAKE builds
+
+LOONGARCH64:
+   - added ABI detection
+   - added support for cpu affinity handling
+   - fixed compilation with early versions of the Loongson toolchain
+   - added an optimized SGEMM kernel for 3A5000
+   - added optimized DGEMV kernels for 3A5000
+   - improved the performance of the DGEMM kernel for 3A5000
+
+MIPS64:
+   - fixed miscompilation of TRMM kernels for the MIPS64_GENERIC target
+
+POWER:
+   - fixed compiler warnings in the POWER10 SBGEMM kernel
+
+RISCV:
+   - fixed application of the INTERFACE64 option when building with CMAKE
+   - fix a potential misdetection of RISCV hardware as 32bit in CMAKE builds
+   - fixed IDAMAX and DOT kernels for C910V
+   - fixed corner cases in the ROT and SWAP kernels for C910V
+   - fixed compilation of the C910V target with recent vendor compilers
+
 ====================================================================
 Version 0.3.23
  01-Apr-2023

Jenkinsfile

@@ -1,9 +1,14 @@
-node {
-        stage('Checkout') {
-            checkout
+pipeline {
+    agent { 
+        docker {
+            image 'osuosl/ubuntu-s390x'
         }
-
+    }
+    stages {
         stage('Build') {
-            sh("make")
+            steps {
+                sh 'make clean && make'
+            }
         }
+    }
 }

Jenkinsfile.pwr

@@ -0,0 +1,16 @@
+pipeline {
+    agent { 
+        docker {
+            image 'osuosl/ubuntu-ppc64le'
+        }
+    }
+    stages {
+        stage('Build') {
+            steps {
+                sh 'sudo apt update'
+                sh 'sudo apt install gfortran -y'
+                sh 'make clean && make'
+            }
+        }
+    }
+}

Makefile

@@ -40,9 +40,9 @@ LAPACK_NOOPT := $(filter-out -O0 -O1 -O2 -O3 -Ofast -O -Og -Os,$(LAPACK_FFLAGS))
 SUBDIRS_ALL = $(SUBDIRS) test ctest utest exports benchmark ../laswp ../bench cpp_thread_test
 
 .PHONY : all libs netlib $(RELA) test ctest shared install
-.NOTPARALLEL : all libs $(RELA) prof lapack-test install blas-test
+.NOTPARALLEL : shared
 
-all :: libs netlib $(RELA) tests shared
+all :: tests
 	@echo
 	@echo " OpenBLAS build complete. ($(LIB_COMPONENTS))"
 	@echo
@@ -150,7 +150,7 @@ ifeq ($(OSNAME), CYGWIN_NT)
 endif
 endif
 
-tests : libs netlib $(RELA) shared
+tests : shared
 ifeq ($(NOFORTRAN), $(filter 0,$(NOFORTRAN)))
 	touch $(LIBNAME)
 ifndef NO_FBLAS
@@ -373,10 +373,10 @@ ifneq ($(CROSS), 1)
 	(cd $(NETLIB_LAPACK_DIR); ./lapack_testing.py -r -b TESTING)
 endif
 
-lapack-runtest:
+lapack-runtest: lapack-test
 	( cd $(NETLIB_LAPACK_DIR)/INSTALL; ./testlsame; ./testslamch; ./testdlamch; \
         ./testsecond; ./testdsecnd; ./testieee; ./testversion )
-	(cd $(NETLIB_LAPACK_DIR); ./lapack_testing.py -r )
+	(cd $(NETLIB_LAPACK_DIR); ./lapack_testing.py -r -b TESTING )
 
 
 blas-test:

Makefile.arm64

@@ -69,7 +69,7 @@ endif
 # in GCC>=9
 ifeq ($(CORE), NEOVERSEN1)
 ifeq (1, $(filter 1,$(GCCVERSIONGTEQ7) $(ISCLANG)))
-ifeq ($(GCCVERSIONGTEQ9), 1)
+ifeq (1, $(filter 1,$(GCCVERSIONGTEQ9) $(ISCLANG)))
 CCOMMON_OPT += -march=armv8.2-a -mtune=neoverse-n1
 ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8.2-a -mtune=neoverse-n1
@@ -92,9 +92,14 @@ endif
 # in GCC>=10.4
 ifeq ($(CORE), NEOVERSEV1)
 ifeq (1, $(filter 1,$(GCCVERSIONGTEQ7) $(ISCLANG)))
-ifeq ($(GCCVERSIONGTEQ10), 1)
-ifeq (1, $(filter 1,$(GCCMINORVERSIONGTEQ4) $(GCCVERSIONGTEQ11)))
-CCOMMON_OPT += -march=armv8.4-a+sve -mtune=neoverse-v1
+ifeq (1, $(filter 1,$(GCCVERSIONGTEQ10) $(ISCLANG)))
+ifeq (1, $(filter 1,$(GCCMINORVERSIONGTEQ4) $(GCCVERSIONGTEQ11) $(ISCLANG)))
+CCOMMON_OPT += -march=armv8.4-a+sve
+ifeq (1, $(ISCLANG))
+CCOMMON_OPT += -mtune=cortex-x1
+else
+CCOMMON_OPT += -mtune=neoverse-v1
+endif
 ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8.4-a -mtune=neoverse-v1
 endif
@@ -122,8 +127,8 @@ endif
 # in GCC>=10.4
 ifeq ($(CORE), NEOVERSEN2)
 ifeq (1, $(filter 1,$(GCCVERSIONGTEQ7) $(ISCLANG)))
-ifeq ($(GCCVERSIONGTEQ10), 1)
-ifeq (1, $(filter 1,$(GCCMINORVERSIONGTEQ4) $(GCCVERSIONGTEQ11)))
+ifeq (1, $(filter 1,$(GCCVERSIONGTEQ10) $(ISCLANG)))
+ifeq (1, $(filter 1,$(GCCMINORVERSIONGTEQ4) $(GCCVERSIONGTEQ11) $(ISCLANG)))
 ifneq ($(OSNAME), Darwin)
 CCOMMON_OPT += -march=armv8.5-a+sve+sve2+bf16 -mtune=neoverse-n2
 else
@@ -155,7 +160,7 @@ endif
 # Use a53 tunings because a55 is only available in GCC>=8.1
 ifeq ($(CORE), CORTEXA55)
 ifeq (1, $(filter 1,$(GCCVERSIONGTEQ7) $(ISCLANG)))
-ifeq ($(GCCVERSIONGTEQ8), 1)
+ifeq (1, $(filter 1,$(GCCVERSIONGTEQ8) $(ISCLANG)))
 CCOMMON_OPT += -march=armv8.2-a -mtune=cortex-a55
 ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8.2-a -mtune=cortex-a55
@@ -196,8 +201,13 @@ endif
 endif
 
 ifeq ($(CORE), THUNDERX3T110)
-ifeq ($(GCCVERSIONGTEQ10), 1)
-CCOMMON_OPT += -march=armv8.3-a -mtune=thunderx3t110
+ifeq (1, $(filter 1,$(GCCVERSIONGTEQ10) $(ISCLANG)))
+CCOMMON_OPT += -march=armv8.3-a 
+ifeq (0, $(ISCLANG))
+CCOMMON_OPT += -mtune=thunderx3t110
+else
+CCOMMON_OPT += -mtune=thunderx2t99
+endif
 ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8.3-a -mtune=thunderx3t110
 endif
@@ -225,9 +235,12 @@ endif
 endif
 endif
 
-ifeq ($(GCCVERSIONGTEQ9), 1)
+ifeq (1, $(filter 1,$(GCCVERSIONGTEQ9) $(ISCLANG)))
 ifeq ($(CORE), EMAG8180)
-CCOMMON_OPT += -march=armv8-a -mtune=emag
+CCOMMON_OPT += -march=armv8-a
+ifeq  ($(ISCLANG), 0)
+CCOMMON_OPT += -mtune=emag
+endif
 ifneq ($(F_COMPILER), NAG)
 FCOMMON_OPT += -march=armv8-a -mtune=emag
 endif

Makefile.rule

@@ -3,7 +3,7 @@
 #
 
 # This library's version
-VERSION = 0.3.23
+VERSION = 0.3.23.dev
 
 # If you set the suffix, the library name will be libopenblas_$(LIBNAMESUFFIX).a
 # and libopenblas_$(LIBNAMESUFFIX).so. Meanwhile, the soname in shared library

Makefile.system

@@ -384,6 +384,11 @@ GCCMINORVERSIONGTEQ4 := $(shell expr `$(CC) $(GCCDUMPVERSION_PARAM) | cut -f2 -d
 GCCMINORVERSIONGTEQ7 := $(shell expr `$(CC) $(GCCDUMPVERSION_PARAM) | cut -f2 -d.` \>= 7)
 endif
 
+ifeq ($(C_COMPILER), CLANG)
+CLANGVERSIONGTEQ9 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 9)
+CLANGVERSIONGTEQ12 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 12)
+endif
+
 #
 #  OS dependent settings
 #
@@ -645,7 +650,7 @@ DYNAMIC_CORE += HASWELL ZEN
 endif
 ifneq ($(NO_AVX512), 1)
 ifneq ($(NO_AVX2), 1)
-DYNAMIC_CORE += SKYLAKEX COOPERLAKE
+DYNAMIC_CORE += SKYLAKEX COOPERLAKE SAPPHIRERAPIDS
 endif
 endif
 endif
@@ -668,6 +673,7 @@ DYNAMIC_CORE += NEOVERSEN1
 ifneq ($(NO_SVE), 1)
 DYNAMIC_CORE += NEOVERSEV1
 DYNAMIC_CORE += NEOVERSEN2
+DYNAMIC_CORE += ARMV8SVE
 endif
 DYNAMIC_CORE += CORTEXA55
 DYNAMIC_CORE += FALKOR
@@ -932,8 +938,12 @@ BINARY_DEFINED = 1
 endif
 
 ifeq ($(ARCH), loongarch64)
-CCOMMON_OPT += -march=loongarch64 -mabi=lp64
-FCOMMON_OPT += -march=loongarch64 -mabi=lp64
+LA64_ABI=$(shell $(CC) -mabi=lp64d -c $(TOPDIR)/cpuid_loongarch64.c -o /dev/null > /dev/null 2> /dev/null && echo lp64d)
+ifneq ($(LA64_ABI), lp64d)
+LA64_ABI=lp64
+endif
+CCOMMON_OPT += -march=loongarch64 -mabi=$(LA64_ABI)
+FCOMMON_OPT += -march=loongarch64 -mabi=$(LA64_ABI)
 endif
 
 endif
@@ -1082,8 +1092,9 @@ endif
 endif
 endif
 
-ifeq ($(F_COMPILER), GFORTRAN)
+ifeq ($(F_COMPILER), $(filter $(F_COMPILER),GFORTRAN FLANGNEW))
 CCOMMON_OPT += -DF_INTERFACE_GFORT
+ifeq ($(F_COMPILER), GFORTRAN)
 FCOMMON_OPT += -Wall
 # make single-threaded LAPACK calls thread-safe #1847
 FCOMMON_OPT += -frecursive
@@ -1097,6 +1108,7 @@ EXTRALIB += -lgfortran
 endif
 endif
 endif
+endif
 ifdef NO_BINARY_MODE
 ifeq ($(ARCH), $(filter $(ARCH),mips64))
 ifdef BINARY64
@@ -1763,6 +1775,8 @@ export TARGET_CORE
 export NO_AVX512
 export NO_AVX2
 export BUILD_BFLOAT16
+export NO_LSX
+export NO_LASX
 
 export SBGEMM_UNROLL_M
 export SBGEMM_UNROLL_N

Makefile.x86_64

@@ -75,18 +75,31 @@ endif
 ifeq ($(CORE), COOPERLAKE)
 ifndef NO_AVX512
 ifeq ($(C_COMPILER), GCC)
-# cooperlake support was added in 10.1
-ifeq ($(GCCVERSIONGTEQ10)$(GCCMINORVERSIONGTEQ1), 11)
-CCOMMON_OPT += -march=cooperlake
-ifneq ($(F_COMPILER), NAG)
-FCOMMON_OPT += -march=cooperlake
-endif
-else  # gcc not support, fallback to avx512
-CCOMMON_OPT += -march=skylake-avx512
-ifneq ($(F_COMPILER), NAG)
-FCOMMON_OPT += -march=skylake-avx512
-endif
-endif
+ # cooperlake support was added in 10.1
+ ifeq ($(GCCVERSIONGTEQ10)$(GCCMINORVERSIONGTEQ1), 11)
+  CCOMMON_OPT += -march=cooperlake
+  ifneq ($(F_COMPILER), NAG)
+   FCOMMON_OPT += -march=cooperlake
+  endif
+ else  # gcc not support, fallback to avx512
+  CCOMMON_OPT += -march=skylake-avx512
+  ifneq ($(F_COMPILER), NAG)
+   FCOMMON_OPT += -march=skylake-avx512
+  endif
+ endif
+else ifeq ($(C_COMPILER), CLANG)
+ # cooperlake support was added in clang 9
+ ifeq ($(CLANGVERSIONGTEQ9), 1)
+  CCOMMON_OPT += -march=cooperlake
+  ifneq ($(F_COMPILER), NAG)
+   FCOMMON_OPT += -march=cooperlake
+  endif
+ else  # not supported in clang, fallback to avx512
+  CCOMMON_OPT += -march=skylake-avx512
+  ifneq ($(F_COMPILER), NAG)
+   FCOMMON_OPT += -march=skylake-avx512
+  endif
+ endif
 endif
 ifeq ($(OSNAME), CYGWIN_NT)
 CCOMMON_OPT += -fno-asynchronous-unwind-tables
@@ -104,18 +117,31 @@ endif
 ifeq ($(CORE), SAPPHIRERAPIDS)
 ifndef NO_AVX512
 ifeq ($(C_COMPILER), GCC)
-# sapphire rapids support was added in 11
-ifeq ($(GCCVERSIONGTEQ11), 1)
-CCOMMON_OPT += -march=sapphirerapids
-ifneq ($(F_COMPILER), NAG)
-FCOMMON_OPT += -march=sapphirerapids
-endif
-else  # gcc not support, fallback to avx512
-CCOMMON_OPT += -march=skylake-avx512
-ifneq ($(F_COMPILER), NAG)
-FCOMMON_OPT += -march=skylake-avx512
-endif
-endif
+ # sapphire rapids support was added in 11
+ ifeq ($(GCCVERSIONGTEQ11), 1)
+  CCOMMON_OPT += -march=sapphirerapids
+  ifneq ($(F_COMPILER), NAG)
+   FCOMMON_OPT += -march=sapphirerapids
+  endif
+ else  # gcc not support, fallback to avx512
+  CCOMMON_OPT += -march=skylake-avx512
+  ifneq ($(F_COMPILER), NAG)
+   FCOMMON_OPT += -march=skylake-avx512
+  endif
+ endif
+else ifeq ($(C_COMPILER), CLANG)
+ # cooperlake support was added in clang 12
+ ifeq ($(CLANGVERSIONGTEQ12), 1)
+  CCOMMON_OPT += -march=cooperlake
+  ifneq ($(F_COMPILER), NAG)
+   FCOMMON_OPT += -march=cooperlake
+  endif
+ else  # not supported in clang, fallback to avx512
+  CCOMMON_OPT += -march=skylake-avx512
+  ifneq ($(F_COMPILER), NAG)
+   FCOMMON_OPT += -march=skylake-avx512
+  endif
+ endif
 endif
 ifeq ($(OSNAME), CYGWIN_NT)
 CCOMMON_OPT += -fno-asynchronous-unwind-tables

README.md

@@ -6,11 +6,15 @@ Travis CI: [![Build Status](https://travis-ci.com/xianyi/OpenBLAS.svg?branch=dev
 
 AppVeyor: [![Build status](https://ci.appveyor.com/api/projects/status/09sohd35n8nkkx64/branch/develop?svg=true)](https://ci.appveyor.com/project/xianyi/openblas/branch/develop)
 
-Drone CI: [![Build Status](https://cloud.drone.io/api/badges/xianyi/OpenBLAS/status.svg?branch=develop)](https://cloud.drone.io/xianyi/OpenBLAS/)
+Cirrus CI: [![Build Status](https://api.cirrus-ci.com/github/xianyi/OpenBLAS.svg?branch=develop)](https://cirrus-ci.com/github/xianyi/OpenBLAS)
+<!-- Drone CI: [![Build Status](https://cloud.drone.io/api/badges/xianyi/OpenBLAS/status.svg?branch=develop)](https://cloud.drone.io/xianyi/OpenBLAS/)-->
+
 
 [![Build Status](https://dev.azure.com/xianyi/OpenBLAS/_apis/build/status/xianyi.OpenBLAS?branchName=develop)](https://dev.azure.com/xianyi/OpenBLAS/_build/latest?definitionId=1&branchName=develop)
 
+OSUOSL POWERCI [![Build Status](https://powerci.osuosl.org/buildStatus/icon?job=OpenBLAS_gh%2Fdevelop)](http://powerci.osuosl.org/job/OpenBLAS_gh/job/develop/)
 
+OSUOSL IBMZ-CI [![Build Status](http://ibmz-ci.osuosl.org/buildStatus/icon?job=OpenBLAS-Z%2Fdevelop)](http://ibmz-ci.osuosl.org/job/OpenBLAS-Z/job/develop/)
 ## Introduction
 
 OpenBLAS is an optimized BLAS (Basic Linear Algebra Subprograms) library based on GotoBLAS2 1.13 BSD version.

azure-pipelines.yml

@@ -115,7 +115,7 @@ jobs:
       mkdir build 
       cd build
       call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat"
-      cmake -G "Ninja" -DCMAKE_C_COMPILER=clang-cl -DCMAKE_CXX_COMPILER=clang-cl -DCMAKE_Fortran_COMPILER=flang -DBUILD_TESTING=OFF -DCMAKE_MT=mt -DCMAKE_BUILD_TYPE=Release -DMSVC_STATIC_CRT=ON ..
+      cmake -G "Ninja" -DCMAKE_C_COMPILER=clang-cl -DCMAKE_CXX_COMPILER=clang-cl -DCMAKE_Fortran_COMPILER="flang -I C:\Miniconda\Library\include\flang" -DBUILD_TESTING=OFF -DCMAKE_MT=mt -DCMAKE_BUILD_TYPE=Release -DMSVC_STATIC_CRT=ON ..
       cmake --build . --config Release
       ctest
 
@@ -271,6 +271,19 @@ jobs:
   - script: |
      make TARGET=ARMV7 DYNAMIC_ARCH=1 NUM_THREADS=32 HOSTCC=clang NOFORTRAN=1
 
+- job: OSX_xbuild_DYNAMIC_ARM64
+  pool:
+     vmImage: 'macOS-11'
+  variables:
+     CC: /Applications/Xcode_12.5.1.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang
+     CFLAGS: -O2 -Wno-macro-redefined -isysroot /Applications/Xcode_12.5.1.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX11.3.sdk -arch arm64
+  steps:
+  - script: |
+     ls /Applications/Xcode_12.5.1.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs
+     /Applications/Xcode_12.5.1.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang -arch arm64 --print-supported-cpus
+     /Applications/Xcode_11.7.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang --version
+     make TARGET=ARMV8 DYNAMIC_ARCH=1 NUM_THREADS=32 HOSTCC=clang NOFORTRAN=1
+
 - job: ALPINE_MUSL
   pool:
      vmImage: 'ubuntu-latest'

benchmark/syrk.c

@@ -1,5 +1,5 @@
 /***************************************************************************
-Copyright (c) 2014, The OpenBLAS Project
+Copyright (c) 2014, 2023 The OpenBLAS Project
 All rights reserved.
 Redistribution and use in source and binary forms, with or without
 modification, are permitted provided that the following conditions are
@@ -67,7 +67,7 @@ int main(int argc, char *argv[]){
   int step =   1;
   int loops =  1;
 
-  if ((p = getenv("OPENBLAS_LOOPS"))) loops=*p;
+  if ((p = getenv("OPENBLAS_LOOPS"))) loops=atoi(p);
 
   double time1,timeg;
 
@@ -77,7 +77,7 @@ int main(int argc, char *argv[]){
   if (argc > 0) { to       = MAX(atol(*argv), from);	argc--; argv++;}
   if (argc > 0) { step     = atol(*argv);		argc--; argv++;}
 
-  fprintf(stderr, "From : %3d  To : %3d Step = %3d Uplo = %c Trans = %c\n", from, to, step,uplo,trans);
+  fprintf(stderr, "From : %3d  To : %3d Step = %3d Uplo = %c Trans = %c Loops = %d\n", from, to, step,uplo,trans,loops);
 
 
   if (( a = (FLOAT *)malloc(sizeof(FLOAT) * to * to * COMPSIZE)) == NULL){

c_check

@@ -31,13 +31,17 @@ flags="$*"
 
 cross_suffix=""
 
-if [ "`dirname \"$compiler_name\"`" != '.' ]; then
-    cross_suffix="$cross_suffix`dirname \"$compiler_name\"`/"
+if [ "`dirname "$compiler_name"`" != '.' ]; then
+    cross_suffix="$cross_suffix`dirname "$compiler_name"`/"
 fi
 
-bn=`basename $compiler_name`
+cn=`echo $compiler_name | sed -e 's/ -.*//'`
+bn=`basename "$cn"`
+
 case "$bn" in
-    *-*) cross_suffix="$cross_suffix${bn%-*}-"
+    *-*) if [ "$bn" != '-' ]; then
+           cross_suffix="$cross_suffix${bn%-*}-"
+         fi
 esac
 
 compiler=""
@@ -164,7 +168,7 @@ fi
 
 no_msa=0
 if [ "$architecture" = "mips" ] || [ "$architecture" = "mips64" ]; then
-    tmpd="$(mktemp -d)"
+    tmpd=$(mktemp -d 2>/dev/null || mktemp -d -t 'OBC')
     tmpf="$tmpd/a.c"
     code='"addvi.b $w0, $w1, 1"'
     msa_flags='-mmsa -mfp64 -mload-store-pairs'
@@ -181,6 +185,37 @@ if [ "$architecture" = "mips" ] || [ "$architecture" = "mips64" ]; then
     rm -rf "$tmpd"
 fi
 
+no_lsx=0
+no_lasx=0
+if [ "$architecture" = "loongarch64" ]; then
+    tmpd="$(mktemp -d)"
+    tmplsx="$tmpd/lsx.c"
+    codelsx='"vadd.b $vr0, $vr0, $vr0"'
+    lsx_flags='-march=loongarch64 -mlsx'
+    printf "#include <lsxintrin.h>\n\n" >> "$tmplsx"
+    printf "void main(void){ __asm__ volatile(%s);}\n" "$codelsx" >> "$tmplsx"
+    args="$lsx_flags -o $tmplsx.o $tmplsx"
+    {
+        $compiler_name $flags $args >/dev/null 2>&1
+    } || {
+        no_lsx=1
+    }
+
+    tmplasx="$tmpd/lasx.c"
+    codelasx='"xvadd.b $xr0, $xr0, $xr0"'
+    lasx_flags='-march=loongarch64 -mlasx'
+    printf "#include <lasxintrin.h>\n\n" >> "$tmplasx"
+    printf "void main(void){ __asm__ volatile(%s);}\n" "$codelasx" >> "$tmplasx"
+    args="$lasx_flags -o $tmplasx.o $tmplasx"
+    {
+        $compiler_name $flags $args >/dev/null 2>&1
+    } || {
+        no_lasx=1
+    }
+
+    rm -rf "$tmpd"
+fi
+
 case "$data" in
     *ARCH_X86_64*) architecture=x86_64 ;;
     *ARCH_X86*) architecture=x86 ;;
@@ -204,7 +239,7 @@ esac
 
 no_avx512=0
 if [ "$architecture" = "x86" ] || [ "$architecture" = "x86_64" ]; then
-    tmpd=`mktemp -d`
+    tmpd=$(mktemp -d 2>/dev/null || mktemp -d -t 'OBC')
     tmpf="$tmpd/a.c"
     code='"vbroadcastss -4 * 4(%rsi), %zmm2"'
     printf "#include <immintrin.h>\n\nint main(void){ __asm__ volatile(%s); }\n" "$code" >> "$tmpf"
@@ -225,7 +260,7 @@ fi
 
 no_rv64gv=0
 if [ "$architecture" = "riscv64" ]; then
-    tmpd=`mktemp -d`
+    tmpd=$(mktemp -d 2>/dev/null || mktemp -d -t 'OBC')
     tmpf="$tmpd/a.c"
     code='"vsetvli    zero, zero, e8, m1\n"'
     printf "int main(void){ __asm__ volatile(%s); }\n" "$code" >> "$tmpf"
@@ -241,13 +276,16 @@ fi
 
 no_sve=0
 if [ "$architecture" = "arm64" ]; then
-    tmpd=`mktemp -d`
+    tmpd=$(mktemp -d 2>/dev/null || mktemp -d -t 'OBC')
     tmpf="$tmpd/a.c"
     printf "#include <arm_sve.h>\n\n int main(void){}\n">> "$tmpf"
     args=" -march=armv8-a+sve -c -o $tmpf.o $tmpf"
     no_sve=0
     {
         $compiler_name $flags $args >/dev/null 2>&1
+    } || {
+        args=" -Msve_intrinsics -c -o $tmpf.o $tmpf"
+        $compiler_name $flags $args >/dev/null 2>&1
     } || {
         no_sve=1
     }
@@ -257,7 +295,7 @@ fi
 c11_atomics=0
 case "$data" in
     *HAVE_C11*)
-        tmpd=`mktemp -d`
+        tmpd=$(mktemp -d 2>/dev/null || mktemp -d -t 'OBC')
         tmpf="$tmpd/a.c"
         printf "#include <stdatomic.h>\nint main(void){}\n" >> "$tmpf"
         args=" -c -o $tmpf.o $tmpf"
@@ -395,6 +433,8 @@ done
     [ "$no_avx512" -eq 1 ] && printf "NO_AVX512=1\n"
     [ "$no_avx2" -eq 1 ] && printf "NO_AVX2=1\n"
     [ "$oldgcc" -eq 1 ] && printf "OLDGCC=1\n"
+    [ "$no_lsx" -eq 1 ] && printf "NO_LSX=1\n"
+    [ "$no_lasx" -eq 1 ] && printf "NO_LASX=1\n"
 } >> "$makefile"
 
 os=`echo "$os" | tr '[[:lower:]]' '[[:upper:]]'/ `
@@ -410,6 +450,8 @@ compiler=`echo "$compiler" | tr '[[:lower:]]' '[[:upper:]]' `
     [ -n "$need_fu" ] && printf "#define FUNDERSCORE\t%s\n" "$need_fu"
     [ "$no_msa" -eq 1 ] && printf "#define NO_MSA\t1\n"
     [ "$c11_atomics" -eq 1 ] && printf "#define HAVE_C11\t1\n"
+    [ "$no_lsx" -eq 1 ] && printf "#define NO_LSX\t1\n"
+    [ "$no_lasx" -eq 1 ] && printf "#define NO_LASX\t1\n"
 } >> "$config"
 
 

c_check.pl

@@ -232,6 +232,47 @@ if (($architecture eq "mips") || ($architecture eq "mips64")) {
     }
 }
 
+$no_lsx = 0;
+$no_lasx = 0;
+if (($architecture eq "loongarch64")) {
+    eval "use File::Temp qw(tempfile)";
+    if ($@){
+	warn "could not load PERL module File::Temp, so could not check LSX and LASX capatibility";
+    } else {
+	$tmplsx = new File::Temp( SUFFIX => '.c' , UNLINK => 1 );
+	$codelsx = '"vadd.b $vr0, $vr0, $vr0"';
+	$lsx_flags = "-march=loongarch64 -mlsx";
+	print $tmplsx "#include <lsxintrin.h>\n\n";
+	print $tmplsx "void main(void){ __asm__ volatile($codelsx); }\n";
+
+	$args = "$lsx_flags -o $tmplsx.o $tmplsx";
+	my @cmd = ("$compiler_name $flags $args >/dev/null 2>/dev/null");
+	system(@cmd) == 0;
+	if ($? != 0) {
+	    $no_lsx = 1;
+	} else {
+	    $no_lsx = 0;
+	}
+	unlink("$tmplsx.o");
+
+	$tmplasx = new File::Temp( SUFFIX => '.c' , UNLINK => 1 );
+	$codelasx = '"xvadd.b $xr0, $xr0, $xr0"';
+	$lasx_flags = "-march=loongarch64 -mlasx";
+	print $tmplasx "#include <lasxintrin.h>\n\n";
+	print $tmplasx "void main(void){ __asm__ volatile($codelasx); }\n";
+
+	$args = "$lasx_flags -o $tmplasx.o $tmplasx";
+	my @cmd = ("$compiler_name $flags $args >/dev/null 2>/dev/null");
+	system(@cmd) == 0;
+	if ($? != 0) {
+	    $no_lasx = 1;
+	} else {
+	    $no_lasx = 0;
+	}
+	unlink("$tmplasx.o");
+    }
+}
+
 $architecture = x86          if ($data =~ /ARCH_X86/);
 $architecture = x86_64       if ($data =~ /ARCH_X86_64/);
 $architecture = e2k          if ($data =~ /ARCH_E2K/);
@@ -424,6 +465,8 @@ print MAKEFILE "NO_RV64GV=1\n" if $no_rv64gv eq 1;
 print MAKEFILE "NO_AVX512=1\n" if $no_avx512 eq 1;
 print MAKEFILE "NO_AVX2=1\n" if $no_avx2 eq 1;
 print MAKEFILE "OLDGCC=1\n" if $oldgcc eq 1;
+print MAKEFILE "NO_LSX=1\n" if $no_lsx eq 1;
+print MAKEFILE "NO_LASX=1\n" if $no_lasx eq 1;
 
 $os           =~ tr/[a-z]/[A-Z]/;
 $architecture =~ tr/[a-z]/[A-Z]/;
@@ -437,6 +480,8 @@ print CONFFILE "#define __64BIT__\t1\n"  if $binformat eq bin64;
 print CONFFILE "#define FUNDERSCORE\t$need_fu\n" if $need_fu ne "";
 print CONFFILE "#define HAVE_MSA\t1\n"  if $have_msa eq 1;
 print CONFFILE "#define HAVE_C11\t1\n" if $c11_atomics eq 1;
+print CONFFILE "#define NO_LSX\t1\n" if $no_lsx eq 1;
+print CONFFILE "#define NO_LASX\t1\n" if $no_lasx eq 1;
 
 
 if ($os eq "LINUX") {

cblas.h

@@ -350,7 +350,7 @@ void cblas_cher2k(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBL
 void cblas_zher2k(OPENBLAS_CONST enum CBLAS_ORDER Order, OPENBLAS_CONST enum CBLAS_UPLO Uplo, OPENBLAS_CONST enum CBLAS_TRANSPOSE Trans, OPENBLAS_CONST blasint N, OPENBLAS_CONST blasint K,
                   OPENBLAS_CONST void *alpha, OPENBLAS_CONST void *A, OPENBLAS_CONST blasint lda, OPENBLAS_CONST void *B, OPENBLAS_CONST blasint ldb, OPENBLAS_CONST double beta, void *C, OPENBLAS_CONST blasint ldc);
 
-void cblas_xerbla(blasint p, char *rout, char *form, ...);
+void cblas_xerbla(blasint p, OPENBLAS_CONST char *rout, OPENBLAS_CONST char *form, ...);
 
 /*** BLAS extensions ***/
 

cmake/arch.cmake

@@ -46,7 +46,7 @@ if (DYNAMIC_ARCH)
   if (ARM64)
 	  set(DYNAMIC_CORE ARMV8 CORTEXA53 CORTEXA55 CORTEXA57 CORTEXA72 CORTEXA73 FALKOR THUNDERX THUNDERX2T99 TSV110 EMAG8180 NEOVERSEN1 THUNDERX3T110)
     if (${CMAKE_C_COMPILER_VERSION} VERSION_GREATER 9.99)
-          set(DYNAMIC_CORE "${DYNAMIC_CORE} NEOVERSEV1 NEOVERSEN2")
+          set(DYNAMIC_CORE ${DYNAMIC_CORE} NEOVERSEV1 NEOVERSEN2 ARMV8SVE)
     endif ()
     if (DYNAMIC_LIST)
 	  set(DYNAMIC_CORE ARMV8 ${DYNAMIC_LIST})
@@ -82,7 +82,7 @@ if (DYNAMIC_ARCH)
       set(DYNAMIC_CORE ${DYNAMIC_CORE} HASWELL ZEN)
     endif ()
     if (NOT NO_AVX512)
-      set(DYNAMIC_CORE ${DYNAMIC_CORE} SKYLAKEX COOPERLAKE)
+      set(DYNAMIC_CORE ${DYNAMIC_CORE} SKYLAKEX COOPERLAKE SAPPHIRERAPIDS)
       string(REGEX REPLACE "-march=native" "" CMAKE_C_FLAGS "${CMAKE_C_FLAGS}")
     endif ()
     if (DYNAMIC_LIST)
@@ -135,7 +135,7 @@ if (ARM64)
   set(BINARY_DEFINED 1)
 endif ()
 
-if (${ARCH} STREQUAL "riscv64")
+if (RISCV64)
   set(NO_BINARY_MODE 1)
   set(BINARY_DEFINED 1)
 endif ()

cmake/cc.cmake

@@ -65,6 +65,14 @@ if (${CMAKE_C_COMPILER_ID} STREQUAL "PGI")
   endif ()
 endif ()
 
+if (${CMAKE_C_COMPILER_ID} STREQUAL "NVHPC")
+  if (POWER)
+    set(CCOMMON_OPT "${CCOMMON_OPT} -tp pwr8")
+  else ()
+    set(CCOMMON_OPT "${CCOMMON_OPT} -tp px")
+  endif ()
+endif ()
+
 if (${CMAKE_C_COMPILER_ID} STREQUAL "PATHSCALE")
   if (BINARY64)
     set(CCOMMON_OPT "${CCOMMON_OPT} -m64")
@@ -172,22 +180,30 @@ endif ()
 
 if (${CORE} STREQUAL NEOVERSEN2)
   if (NOT DYNAMIC_ARCH)
-    execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpversion OUTPUT_VARIABLE GCC_VERSION)
-    if (${GCC_VERSION} VERSION_GREATER 10.4 OR ${GCC_VERSION} VERSION_EQUAL 10.4)
-      set (CCOMMON_OPT  "${CCOMMON_OPT} -march=armv8.5-a+sve+sve2+bf16 -mtune=neoverse-n2")
+    if (${CMAKE_C_COMPILER_ID} STREQUAL "PGI" AND NOT NO_SVE)
+	set (CCOMMON_OPT  "${CCOMMON_OPT} -Msve_intrinsics -march=armv8.5-a+sve+sve2+bf16 -mtune=neoverse-n2")
     else ()
-      set (CCOMMON_OPT "${CCOMMON_OPT} -march=armv8.2-a+sve")
-    endif()
+      execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpversion OUTPUT_VARIABLE GCC_VERSION)
+      if (${GCC_VERSION} VERSION_GREATER 10.4 OR ${GCC_VERSION} VERSION_EQUAL 10.4)
+	set (CCOMMON_OPT  "${CCOMMON_OPT} -march=armv8.5-a+sve+sve2+bf16 -mtune=neoverse-n2")
+      else ()
+	set (CCOMMON_OPT "${CCOMMON_OPT} -march=armv8.2-a+sve")
+      endif()
+    endif ()
   endif ()
 endif ()
 
 if (${CORE} STREQUAL NEOVERSEV1)
   if (NOT DYNAMIC_ARCH)
-    execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpversion OUTPUT_VARIABLE GCC_VERSION)
-    if (${GCC_VERSION} VERSION_GREATER 10.4 OR ${GCC_VERSION} VERSION_EQUAL 10.4)
-      set (CCOMMON_OPT  "${CCOMMON_OPT} -march=armv8.4-a+sve -mtune=neoverse-v1")
+    if (${CMAKE_C_COMPILER_ID} STREQUAL "PGI" AND NOT NO_SVE)
+	set (CCOMMON_OPT  "${CCOMMON_OPT} -Msve_intrinsics -march=armv8.4-a+sve -mtune=neoverse-v1")
     else ()
-      set (CCOMMON_OPT "${CCOMMON_OPT} -march=armv8.2-a+sve")
+      execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpversion OUTPUT_VARIABLE GCC_VERSION)
+      if (${GCC_VERSION} VERSION_GREATER 10.4 OR ${GCC_VERSION} VERSION_EQUAL 10.4)
+        set (CCOMMON_OPT  "${CCOMMON_OPT} -march=armv8.4-a+sve -mtune=neoverse-v1")
+      else ()
+        set (CCOMMON_OPT "${CCOMMON_OPT} -march=armv8.2-a+sve")
+      endif()
     endif()
   endif ()
 endif ()
@@ -205,7 +221,11 @@ endif ()
 
 if (${CORE} STREQUAL ARMV8SVE)
   if (NOT DYNAMIC_ARCH)
-    set (CCOMMON_OPT "${CCOMMON_OPT} -march=armv8-a+sve")
+    if (${CMAKE_C_COMPILER_ID} STREQUAL "PGI" AND NOT NO_SVE)
+	set (CCOMMON_OPT  "${CCOMMON_OPT} -Msve_intrinsics -march=armv8-a+sve")
+    else ()
+      set (CCOMMON_OPT "${CCOMMON_OPT} -march=armv8-a+sve")
+    endif ()
   endif ()
 endif ()
 

cmake/fc.cmake

@@ -3,7 +3,8 @@
 ## Description: Ported from portion of OpenBLAS/Makefile.system
 ##              Sets Fortran related variables.
 
-if (${F_COMPILER} STREQUAL "FLANG")
+if (${F_COMPILER} STREQUAL "FLANG" AND NOT CMAKE_Fortran_COMPILER_ID STREQUAL "LLVMFlang")
+  # This is for classic Flang. LLVM Flang is handled with gfortran below.
   set(CCOMMON_OPT "${CCOMMON_OPT} -DF_INTERFACE_FLANG")
   if (BINARY64 AND INTERFACE64)
     set(FCOMMON_OPT "${FCOMMON_OPT} -i8")
@@ -38,15 +39,17 @@ if (${F_COMPILER} STREQUAL "G95")
   endif ()
 endif ()
 
-if (${F_COMPILER} STREQUAL "GFORTRAN")
+if (${F_COMPILER} STREQUAL "GFORTRAN" OR ${F_COMPILER} STREQUAL "F95" OR CMAKE_Fortran_COMPILER_ID STREQUAL "LLVMFlang")
   set(CCOMMON_OPT "${CCOMMON_OPT} -DF_INTERFACE_GFORT")
-  # ensure reentrancy of lapack codes
-  set(FCOMMON_OPT "${FCOMMON_OPT} -Wall -frecursive")
-  # work around ABI violation in passing string arguments from C
-  set(FCOMMON_OPT "${FCOMMON_OPT} -fno-optimize-sibling-calls")
-  #Don't include -lgfortran, when NO_LAPACK=1 or lsbcc
-  if (NOT NO_LAPACK)
-    set(EXTRALIB "${EXTRALIB} -lgfortran")
+  if (NOT CMAKE_Fortran_COMPILER_ID STREQUAL "LLVMFlang")
+    # ensure reentrancy of lapack codes
+    set(FCOMMON_OPT "${FCOMMON_OPT} -Wall -frecursive")
+    # work around ABI violation in passing string arguments from C
+    set(FCOMMON_OPT "${FCOMMON_OPT} -fno-optimize-sibling-calls")
+    if (NOT NO_LAPACK)
+      # Don't include -lgfortran, when NO_LAPACK=1 or lsbcc
+      set(EXTRALIB "${EXTRALIB} -lgfortran")
+    endif ()
   endif ()
   if (NO_BINARY_MODE)
     if (MIPS64)
@@ -63,6 +66,13 @@ if (${F_COMPILER} STREQUAL "GFORTRAN")
         set(FCOMMON_OPT "${FCOMMON_OPT} -mabi=lp32")
       endif ()
     endif ()
+    if (RISCV64)
+      if (BINARY64)
+        if (INTERFACE64)
+          set(FCOMMON_OPT "${FCOMMON_OPT} -fdefault-integer-8")
+        endif ()
+      endif ()
+    endif ()
   else ()
     if (BINARY64)
       set(FCOMMON_OPT "${FCOMMON_OPT} -m64")
@@ -121,7 +131,7 @@ if (${F_COMPILER} STREQUAL "IBM")
   endif ()
 endif ()
 
-if (${F_COMPILER} STREQUAL "PGI")
+if (${F_COMPILER} STREQUAL "PGI" OR ${F_COMPILER} STREQUAL "PGF95")
   set(CCOMMON_OPT "${CCOMMON_OPT} -DF_INTERFACE_PGI")
   set(COMMON_PROF "${COMMON_PROF} -DPGICOMPILER")
   if (BINARY64)

cmake/lapack.cmake

@@ -124,7 +124,7 @@ set(SLASRC
    ssbev_2stage.f ssbevx_2stage.f ssbevd_2stage.f ssygv_2stage.f
    sgesvdq.f slaorhr_col_getrfnp.f
    slaorhr_col_getrfnp2.f sorgtsqr.f sorgtsqr_row.f sorhr_col.f 
-   slatrs3.f strsyl3.f sgelst.f)
+   slatrs3.f strsyl3.f sgelst.f sgedmd.f90 sgedmdq.f90)
 
 set(SXLASRC sgesvxx.f sgerfsx.f sla_gerfsx_extended.f sla_geamv.f
    sla_gercond.f sla_gerpvgrw.f ssysvxx.f ssyrfsx.f
@@ -187,7 +187,7 @@ set(CLASRC
    cposv.f  cposvx.f cpotrf2.f cpotri.f cpstrf.f cpstf2.f
    cppcon.f cppequ.f cpprfs.f cppsv.f  cppsvx.f cpptrf.f cpptri.f cpptrs.f
    cptcon.f cpteqr.f cptrfs.f cptsv.f  cptsvx.f cpttrf.f cpttrs.f cptts2.f
-   crot.f   cspcon.f csprfs.f cspsv.f
+   crot.f crscl.f cspcon.f csprfs.f cspsv.f
    cspsvx.f csptrf.f csptri.f csptrs.f csrscl.f cstedc.f
    cstegr.f cstein.f csteqr.f csycon.f
    csyrfs.f csysv.f  csysvx.f csytf2.f csytrf.f csytri.f
@@ -223,7 +223,7 @@ set(CLASRC
    chbev_2stage.f chbevx_2stage.f chbevd_2stage.f chegv_2stage.f
    cgesvdq.f claunhr_col_getrfnp.f claunhr_col_getrfnp2.f 
    cungtsqr.f cungtsqr_row.f cunhr_col.f 
-   clatrs3.f ctrsyl3.f cgelst.f)
+   clatrs3.f ctrsyl3.f cgelst.f cgedmd.f90 cgedmdq.f90)
 
 set(CXLASRC cgesvxx.f cgerfsx.f cla_gerfsx_extended.f cla_geamv.f
    cla_gercond_c.f cla_gercond_x.f cla_gerpvgrw.f
@@ -316,7 +316,7 @@ set(DLASRC
    dsbev_2stage.f dsbevx_2stage.f dsbevd_2stage.f dsygv_2stage.f
    dcombssq.f dgesvdq.f dlaorhr_col_getrfnp.f
    dlaorhr_col_getrfnp2.f dorgtsqr.f dorgtsqr_row.f dorhr_col.f 
-   dlatrs3.f dtrsyl3.f dgelst.f)
+   dlatrs3.f dtrsyl3.f dgelst.f dgedmd.f90 dgedmdq.f90)
 
 set(DXLASRC dgesvxx.f dgerfsx.f dla_gerfsx_extended.f dla_geamv.f
    dla_gercond.f dla_gerpvgrw.f dsysvxx.f dsyrfsx.f
@@ -381,7 +381,7 @@ set(ZLASRC
    zposv.f  zposvx.f zpotrf2.f zpotri.f zpotrs.f zpstrf.f zpstf2.f
    zppcon.f zppequ.f zpprfs.f zppsv.f  zppsvx.f zpptrf.f zpptri.f zpptrs.f
    zptcon.f zpteqr.f zptrfs.f zptsv.f  zptsvx.f zpttrf.f zpttrs.f zptts2.f
-   zrot.f   zspcon.f zsprfs.f zspsv.f
+   zrot.f zrscl.f zspcon.f zsprfs.f zspsv.f
    zspsvx.f zsptrf.f zsptri.f zsptrs.f zdrscl.f zstedc.f
    zstegr.f zstein.f zsteqr.f zsycon.f
    zsyrfs.f zsysv.f  zsysvx.f zsytf2.f zsytrf.f zsytri.f
@@ -419,7 +419,7 @@ set(ZLASRC
    zhbev_2stage.f zhbevx_2stage.f zhbevd_2stage.f zhegv_2stage.f
    zgesvdq.f zlaunhr_col_getrfnp.f zlaunhr_col_getrfnp2.f
    zungtsqr.f zungtsqr_row.f zunhr_col.f
-   zlatrs3.f ztrsyl3.f zgelst.f)
+   zlatrs3.f ztrsyl3.f zgelst.f zgedmd.f90 zgedmdq.f90)
 
 set(ZXLASRC zgesvxx.f zgerfsx.f zla_gerfsx_extended.f zla_geamv.f
    zla_gercond_c.f zla_gercond_x.f zla_gerpvgrw.f zsysvxx.f zsyrfsx.f
@@ -436,6 +436,7 @@ if(USE_XBLAS)
   set(ALLXOBJ ${SXLASRC} ${DXLASRC} ${CXLASRC} ${ZXLASRC})
 endif()
 
+if(BUILD_LAPACK_DEPRECATED)
 list(APPEND SLASRC DEPRECATED/sgegs.f DEPRECATED/sgegv.f
   DEPRECATED/sgeqpf.f DEPRECATED/sgelsx.f DEPRECATED/sggsvd.f
   DEPRECATED/sggsvp.f DEPRECATED/slahrd.f DEPRECATED/slatzm.f DEPRECATED/stzrqf.f)
@@ -449,6 +450,7 @@ list(APPEND ZLASRC DEPRECATED/zgegs.f DEPRECATED/zgegv.f
   DEPRECATED/zgeqpf.f DEPRECATED/zgelsx.f DEPRECATED/zggsvd.f
   DEPRECATED/zggsvp.f DEPRECATED/zlahrd.f DEPRECATED/zlatzm.f DEPRECATED/ztzrqf.f)
 message(STATUS "Building deprecated routines")
+endif()
 
 set(DSLASRC spotrs.f)
 
@@ -622,7 +624,7 @@ set(SLASRC
    ssbev_2stage.c ssbevx_2stage.c ssbevd_2stage.c ssygv_2stage.c
    sgesvdq.c slaorhr_col_getrfnp.c
    slaorhr_col_getrfnp2.c sorgtsqr.c sorgtsqr_row.c sorhr_col.c 
-   slatrs3.c strsyl3.c sgelst.c)
+   slatrs3.c strsyl3.c sgelst.c sgedmd.c sgedmdq.c)
 
 set(SXLASRC sgesvxx.c sgerfsx.c sla_gerfsx_extended.c sla_geamv.c
    sla_gercond.c sla_gerpvgrw.c ssysvxx.c ssyrfsx.c
@@ -684,7 +686,7 @@ set(CLASRC
    cposv.c  cposvx.c cpotrf2.c cpotri.c cpstrf.c cpstf2.c
    cppcon.c cppequ.c cpprfs.c cppsv.c  cppsvx.c cpptrf.c cpptri.c cpptrs.c
    cptcon.c cpteqr.c cptrfs.c cptsv.c  cptsvx.c cpttrf.c cpttrs.c cptts2.c
-   crot.c   cspcon.c csprfs.c cspsv.c
+   crot.c crscl.c  cspcon.c csprfs.c cspsv.c
    cspsvx.c csptrf.c csptri.c csptrs.c csrscl.c cstedc.c
    cstegr.c cstein.c csteqr.c csycon.c
    csyrfs.c csysv.c  csysvx.c csytf2.c csytrf.c csytri.c
@@ -720,7 +722,7 @@ set(CLASRC
    chbev_2stage.c chbevx_2stage.c chbevd_2stage.c chegv_2stage.c
    cgesvdq.c claunhr_col_getrfnp.c claunhr_col_getrfnp2.c 
    cungtsqr.c cungtsqr_row.c cunhr_col.c 
-   clatrs3.c ctrsyl3.c cgelst.c)
+   clatrs3.c ctrsyl3.c cgelst.c cgedmd.c cgedmdq.c)
 
 set(CXLASRC cgesvxx.c cgerfsx.c cla_gerfsx_extended.c cla_geamv.c
    cla_gercond_c.c cla_gercond_x.c cla_gerpvgrw.c
@@ -812,7 +814,7 @@ set(DLASRC
    dsbev_2stage.c dsbevx_2stage.c dsbevd_2stage.c dsygv_2stage.c
    dcombssq.c dgesvdq.c dlaorhr_col_getrfnp.c
    dlaorhr_col_getrfnp2.c dorgtsqr.c dorgtsqr_row.c dorhr_col.c 
-   dlatrs3.c dtrsyl3.c dgelst.c)
+   dlatrs3.c dtrsyl3.c dgelst.c dgedmd.c dgedmdq.c)
 
 set(DXLASRC dgesvxx.c dgerfsx.c dla_gerfsx_extended.c dla_geamv.c
    dla_gercond.c dla_gerpvgrw.c dsysvxx.c dsyrfsx.c
@@ -876,7 +878,7 @@ set(ZLASRC
    zposv.c  zposvx.c zpotrf2.c zpotri.c zpotrs.c zpstrf.c zpstf2.c
    zppcon.c zppequ.c zpprfs.c zppsv.c  zppsvx.c zpptrf.c zpptri.c zpptrs.c
    zptcon.c zpteqr.c zptrfs.c zptsv.c  zptsvx.c zpttrf.c zpttrs.c zptts2.c
-   zrot.c   zspcon.c zsprfs.c zspsv.c
+   zrot.c zrscl.c  zspcon.c zsprfs.c zspsv.c
    zspsvx.c zsptrf.c zsptri.c zsptrs.c zdrscl.c zstedc.c
    zstegr.c zstein.c zsteqr.c zsycon.c
    zsyrfs.c zsysv.c  zsysvx.c zsytf2.c zsytrf.c zsytri.c
@@ -913,7 +915,8 @@ set(ZLASRC
    zheevd_2stage.c zheev_2stage.c zheevx_2stage.c zheevr_2stage.c
    zhbev_2stage.c zhbevx_2stage.c zhbevd_2stage.c zhegv_2stage.c
    zgesvdq.c zlaunhr_col_getrfnp.c zlaunhr_col_getrfnp2.c
-   zungtsqr.c zungtsqr_row.c zunhr_col.c zlatrs3.c ztrsyl3.c zgelst.c)
+   zungtsqr.c zungtsqr_row.c zunhr_col.c zlatrs3.c ztrsyl3.c zgelst.c
+   zgedmd.c zgedmdq.c)
 
 set(ZXLASRC zgesvxx.c zgerfsx.c zla_gerfsx_extended.c zla_geamv.c
    zla_gercond_c.c zla_gercond_x.c zla_gerpvgrw.c zsysvxx.c zsyrfsx.c
@@ -930,6 +933,7 @@ if(USE_XBLAS)
   set(ALLXOBJ ${SXLASRC} ${DXLASRC} ${CXLASRC} ${ZXLASRC})
 endif()
 
+if(BUILD_LAPACK_DEPRECATED)
 list(APPEND SLASRC DEPRECATED/sgegs.c DEPRECATED/sgegv.c
   DEPRECATED/sgeqpf.c DEPRECATED/sgelsx.c DEPRECATED/sggsvd.c
   DEPRECATED/sggsvp.c DEPRECATED/slahrd.c DEPRECATED/slatzm.c DEPRECATED/stzrqf.c)
@@ -943,6 +947,7 @@ list(APPEND ZLASRC DEPRECATED/zgegs.c DEPRECATED/zgegv.c
   DEPRECATED/zgeqpf.c DEPRECATED/zgelsx.c DEPRECATED/zggsvd.c
   DEPRECATED/zggsvp.c DEPRECATED/zlahrd.c DEPRECATED/zlatzm.c DEPRECATED/ztzrqf.c)
 message(STATUS "Building deprecated routines")
+endif()
 
 set(DSLASRC spotrs.c)
 

cmake/lapacke.cmake

@@ -70,8 +70,6 @@ set(CSRC
   lapacke_cgeqlf_work.c
   lapacke_cgeqp3.c
   lapacke_cgeqp3_work.c
-  lapacke_cgeqpf.c
-  lapacke_cgeqpf_work.c
   lapacke_cgeqr.c
   lapacke_cgeqr_work.c
   lapacke_cgeqr2.c
@@ -92,6 +90,10 @@ set(CSRC
   lapacke_cgerqf_work.c
   lapacke_cgesdd.c
   lapacke_cgesdd_work.c
+  lapacke_cgedmd.c
+  lapacke_cgedmd_work.c
+  lapacke_cgedmdq.c
+  lapacke_cgedmdq_work.c
   lapacke_cgesv.c
   lapacke_cgesv_work.c
   lapacke_cgesvd.c
@@ -144,12 +146,8 @@ set(CSRC
   lapacke_cggqrf_work.c
   lapacke_cggrqf.c
   lapacke_cggrqf_work.c
-  lapacke_cggsvd.c
-  lapacke_cggsvd_work.c
   lapacke_cggsvd3.c
   lapacke_cggsvd3_work.c
-  lapacke_cggsvp.c
-  lapacke_cggsvp_work.c
   lapacke_cggsvp3.c
   lapacke_cggsvp3_work.c
   lapacke_cgtcon.c
@@ -564,6 +562,8 @@ set(CSRC
   lapacke_ctrsna_work.c
   lapacke_ctrsyl.c
   lapacke_ctrsyl_work.c
+  lapacke_ctrsyl3.c
+  lapacke_ctrsyl3_work.c
   lapacke_ctrtri.c
   lapacke_ctrtri_work.c
   lapacke_ctrtrs.c
@@ -596,6 +596,8 @@ set(CSRC
   lapacke_cungtr_work.c
   lapacke_cungtsqr_row.c
   lapacke_cungtsqr_row_work.c
+  lapacke_cunhr_col.c
+  lapacke_cunhr_col_work.c
   lapacke_cunmbr.c
   lapacke_cunmbr_work.c
   lapacke_cunmhr.c
@@ -695,8 +697,6 @@ set(DSRC
   lapacke_dgeqlf_work.c
   lapacke_dgeqp3.c
   lapacke_dgeqp3_work.c
-  lapacke_dgeqpf.c
-  lapacke_dgeqpf_work.c
   lapacke_dgeqr.c
   lapacke_dgeqr_work.c
   lapacke_dgeqr2.c
@@ -717,6 +717,10 @@ set(DSRC
   lapacke_dgerqf_work.c
   lapacke_dgesdd.c
   lapacke_dgesdd_work.c
+  lapacke_dgedmd.c
+  lapacke_dgedmd_work.c
+  lapacke_dgedmdq.c
+  lapacke_dgedmdq_work.c
   lapacke_dgesv.c
   lapacke_dgesv_work.c
   lapacke_dgesvd.c
@@ -771,12 +775,8 @@ set(DSRC
   lapacke_dggqrf_work.c
   lapacke_dggrqf.c
   lapacke_dggrqf_work.c
-  lapacke_dggsvd.c
-  lapacke_dggsvd_work.c
   lapacke_dggsvd3.c
   lapacke_dggsvd3_work.c
-  lapacke_dggsvp.c
-  lapacke_dggsvp_work.c
   lapacke_dggsvp3.c
   lapacke_dggsvp3_work.c
   lapacke_dgtcon.c
@@ -874,6 +874,8 @@ set(DSRC
   lapacke_dorgtr_work.c
   lapacke_dorgtsqr_row.c
   lapacke_dorgtsqr_row_work.c
+  lapacke_dorhr_col.c
+  lapacke_dorhr_col_work.c
   lapacke_dormbr.c
   lapacke_dormbr_work.c
   lapacke_dormhr.c
@@ -1186,6 +1188,8 @@ set(DSRC
   lapacke_dtrsna_work.c
   lapacke_dtrsyl.c
   lapacke_dtrsyl_work.c
+  lapacke_dtrsyl3.c
+  lapacke_dtrsyl3_work.c
   lapacke_dtrtri.c
   lapacke_dtrtri_work.c
   lapacke_dtrtrs.c
@@ -1275,8 +1279,6 @@ set(SSRC
   lapacke_sgeqlf_work.c
   lapacke_sgeqp3.c
   lapacke_sgeqp3_work.c
-  lapacke_sgeqpf.c
-  lapacke_sgeqpf_work.c
   lapacke_sgeqr.c
   lapacke_sgeqr_work.c
   lapacke_sgeqr2.c
@@ -1297,6 +1299,10 @@ set(SSRC
   lapacke_sgerqf_work.c
   lapacke_sgesdd.c
   lapacke_sgesdd_work.c
+  lapacke_sgedmd.c
+  lapacke_sgedmd_work.c
+  lapacke_sgedmdq.c
+  lapacke_sgedmdq_work.c
   lapacke_sgesv.c
   lapacke_sgesv_work.c
   lapacke_sgesvd.c
@@ -1351,12 +1357,8 @@ set(SSRC
   lapacke_sggqrf_work.c
   lapacke_sggrqf.c
   lapacke_sggrqf_work.c
-  lapacke_sggsvd.c
-  lapacke_sggsvd_work.c
   lapacke_sggsvd3.c
   lapacke_sggsvd3_work.c
-  lapacke_sggsvp.c
-  lapacke_sggsvp_work.c
   lapacke_sggsvp3.c
   lapacke_sggsvp3_work.c
   lapacke_sgtcon.c
@@ -1453,6 +1455,8 @@ set(SSRC
   lapacke_sorgtr_work.c
   lapacke_sorgtsqr_row.c
   lapacke_sorgtsqr_row_work.c
+  lapacke_sorhr_col.c
+  lapacke_sorhr_col_work.c
   lapacke_sormbr.c
   lapacke_sormbr_work.c
   lapacke_sormhr.c
@@ -1762,6 +1766,8 @@ set(SSRC
   lapacke_strsna_work.c
   lapacke_strsyl.c
   lapacke_strsyl_work.c
+  lapacke_ctrsyl3.c
+  lapacke_ctrsyl3_work.c
   lapacke_strtri.c
   lapacke_strtri_work.c
   lapacke_strtrs.c
@@ -1849,8 +1855,6 @@ set(ZSRC
   lapacke_zgeqlf_work.c
   lapacke_zgeqp3.c
   lapacke_zgeqp3_work.c
-  lapacke_zgeqpf.c
-  lapacke_zgeqpf_work.c
   lapacke_zgeqr.c
   lapacke_zgeqr_work.c
   lapacke_zgeqr2.c
@@ -1871,6 +1875,10 @@ set(ZSRC
   lapacke_zgerqf_work.c
   lapacke_zgesdd.c
   lapacke_zgesdd_work.c
+  lapacke_zgedmd.c
+  lapacke_zgedmd_work.c
+  lapacke_zgedmdq.c
+  lapacke_zgedmdq_work.c
   lapacke_zgesv.c
   lapacke_zgesv_work.c
   lapacke_zgesvd.c
@@ -1925,12 +1933,8 @@ set(ZSRC
   lapacke_zggqrf_work.c
   lapacke_zggrqf.c
   lapacke_zggrqf_work.c
-  lapacke_zggsvd.c
-  lapacke_zggsvd_work.c
   lapacke_zggsvd3.c
   lapacke_zggsvd3_work.c
-  lapacke_zggsvp.c
-  lapacke_zggsvp_work.c
   lapacke_zggsvp3.c
   lapacke_zggsvp3_work.c
   lapacke_zgtcon.c
@@ -2343,6 +2347,8 @@ set(ZSRC
   lapacke_ztrsna_work.c
   lapacke_ztrsyl.c
   lapacke_ztrsyl_work.c
+  lapacke_ztrsyl3.c
+  lapacke_ztrsyl3_work.c
   lapacke_ztrtri.c
   lapacke_ztrtri_work.c
   lapacke_ztrtrs.c
@@ -2375,6 +2381,8 @@ set(ZSRC
   lapacke_zungtr_work.c
   lapacke_zungtsqr_row.c
   lapacke_zungtsqr_row_work.c
+  lapacke_zunhr_col.c
+  lapacke_zunhr_col_work.c
   lapacke_zunmbr.c
   lapacke_zunmbr_work.c
   lapacke_zunmhr.c
@@ -2401,6 +2409,12 @@ set(ZSRC
   lapacke_csyr_work.c
   lapacke_ilaver.c
 )
+if (BUILD_LAPACK_DEPRECATED)
+set(SRCS $SRCS lapacke_sgeqpf.c lapacke_sgeqpf_work.c lapacke_sggsvd.c lapacke_sggsvd_work.c lapacke_sggsvp.c lapacke_sggsvp_work.c)
+set(SRCD $SRCD lapacke_dgeqpf.c lapacke_dgeqpf_work.c lapacke_dggsvd.c lapacke_dggsvd_work.c lapacke_dggsvp.c lapacke_dggsvp_work.c)
+set(SRCC $SRCC lapacke_cgeqpf.c lapacke_cgeqpf_work.c lapacke_cggsvd.c lapacke_cggsvd_work.c lapacke_cggsvp.c lapacke_cggsvp_work.c)
+set(SRCZ $SRCZ lapacke_zgeqpf.c lapacke_zgeqpf_work.c lapacke_zggsvd.c lapacke_zggsvd_work.c lapacke_zggsvp.c lapacke_zggsvp_work.c)
+endif()
 
 set(SRCX
   lapacke_cgbrfsx.c       lapacke_cporfsx.c       lapacke_dgerfsx.c       lapacke_sgbrfsx.c       lapacke_ssyrfsx.c       lapacke_zherfsx.c

cmake/system.cmake

@@ -55,7 +55,7 @@ if (DEFINED TARGET)
 endif ()
 
 # On x86_64 build getarch with march=native. This is required to detect AVX512 support in getarch.
-if (X86_64 AND NOT ${CMAKE_C_COMPILER_ID} STREQUAL "PGI")
+if (X86_64 AND NOT (${CMAKE_C_COMPILER_ID} STREQUAL "PGI" OR ${CMAKE_C_COMPILER_ID} STREQUAL "NVHPC"))
   set(GETARCH_FLAGS "${GETARCH_FLAGS} -march=native")
 endif ()
 
@@ -280,7 +280,41 @@ if (DEFINED TARGET)
   if (${TARGET} STREQUAL POWER8)
     set (KERNEL_DEFINITIONS  "${KERNEL_DEFINITIONS} -mcpu=power8 -mtune=power8 -mvsx -fno-fast-math")
   endif()
+
+if (${TARGET} STREQUAL NEOVERSEV1)
+    if (${CMAKE_C_COMPILER_ID} STREQUAL "PGI" AND NOT NO_SVE)
+	set (KERNEL_DEFINITIONS  "${KERNEL_DEFINITIONS} -Msve_intrinsics -march=armv8.4-a+sve -mtune=neoverse-v1")
+    else ()
+    execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpversion OUTPUT_VARIABLE GCC_VERSION)
+    if (${GCC_VERSION} VERSION_GREATER 10.4 OR ${GCC_VERSION} VERSION_EQUAL 10.4)
+      set (KERNEL_DEFINITIONS  "${KERNEL_DEFINITIONS} -march=armv8.4-a+sve -mtune=neoverse-v1")
+    else ()
+	    message(FATAL_ERROR "Compiler ${CMAKE_C_COMPILER} ${GCC_VERSION} does not support Neoverse V1.")
+    endif()
+    endif()
+  endif()
+  if (${TARGET} STREQUAL NEOVERSEN2)
+    if (${CMAKE_C_COMPILER_ID} STREQUAL "PGI" AND NOT NO_SVE)
+	set (KERNEL_DEFINITIONS  "${KERNEL_DEFINITIONS} -Msve-intrinsics -march=armv8.5-a+sve+sve2+bf16 -mtune=neoverse-n2")
+    else ()
+    execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpversion OUTPUT_VARIABLE GCC_VERSION)
+    if (${GCC_VERSION} VERSION_GREATER 10.4 OR ${GCC_VERSION} VERSION_EQUAL 10.4)
+      set (KERNEL_DEFINITIONS  "${KERNEL_DEFINITIONS} -march=armv8.5-a+sve+sve2+bf16 -mtune=neoverse-n2")
+    else ()
+	    message(FATAL_ERROR "Compiler $${CMAKE_C_COMPILER} {GCC_VERSION} does not support Neoverse N2.")
+    endif()
+    endif()
+  endif()
+  if (${TARGET} STREQUAL ARMV8SVE)
+    if (${CMAKE_C_COMPILER_ID} STREQUAL "PGI" AND NOT NO_SVE)
+      set (KERNEL_DEFINITIONS  "${KERNEL_DEFINITIONS} -Msve-intrinsics -march=armv8.2-a+sve")
+    else ()
+      set (KERNEL_DEFINITIONS  "${KERNEL_DEFINITIONS} -march=armv8.2-a+sve")
+    endif()
+  endif()
+
 endif()
+
 if (DEFINED BINARY)
   message(STATUS "Compiling a ${BINARY}-bit binary.")
 endif ()

cmake/system_check.cmake

@@ -44,6 +44,8 @@ elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "mips64.*")
   set(MIPS64 1)
 elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "loongarch64.*")
   set(LOONGARCH64 1)
+elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "riscv64.*")
+  set(RISCV64 1)
 elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "amd64.*|x86_64.*|AMD64.*")
   if (NOT BINARY)
     if("${CMAKE_SIZEOF_VOID_P}" EQUAL "8")
@@ -60,7 +62,7 @@ elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "amd64.*|x86_64.*|AMD64.*")
   endif()
 elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "i686.*|i386.*|x86.*|amd64.*|AMD64.*")
   set(X86 1)
-elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(aarch64.*|AARCH64.*|arm64.*|ARM64.*)")
+elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(aarch64.*|AARCH64.*|arm64.*|ARM64.*|armv8.*)")
   if("${CMAKE_SIZEOF_VOID_P}" EQUAL "8")
     set(ARM64 1)
   else()
@@ -107,7 +109,7 @@ else()
 endif ()
 
 if (NOT BINARY)
-  if (X86_64 OR ARM64 OR POWER OR MIPS64 OR LOONGARCH64)
+  if (X86_64 OR ARM64 OR POWER OR MIPS64 OR LOONGARCH64 OR RISCV64)
     set(BINARY 64)
   else ()
     set(BINARY 32)

cmake/utils.cmake

@@ -87,6 +87,15 @@ macro(ParseMakefileVars MAKEFILE_IN)
       #message(STATUS "skipping ${makefile_line}")
       continue ()
     endif ()
+
+    # Example 1: SBGEMM_SMALL_M_PERMIT =
+    # Unset the variable
+    string(REGEX MATCH "([0-9_a-zA-Z]+)[ \t]*=[ \t]*$" line_match "${makefile_line}")
+    if (NOT "${line_match}" STREQUAL "")
+      set(var_name ${CMAKE_MATCH_1})
+      unset(${var_name})
+    endif()
+
     string(REGEX MATCH "([0-9_a-zA-Z]+)[ \t]*=[ \t]*(.+)$" line_match "${makefile_line}")
     if (NOT "${line_match}" STREQUAL "")
       #message(STATUS "match on ${line_match}")

common.h

@@ -525,7 +525,7 @@ static inline unsigned long long rpcc(void){
 #endif // !RPCC_DEFINED
 
 #if !defined(BLAS_LOCK_DEFINED) && defined(__GNUC__)
-static void __inline blas_lock(volatile BLASULONG *address){
+static __inline void blas_lock(volatile BLASULONG *address){
 
   do {
     while (*address) {YIELDING;};

common_alpha.h

@@ -45,7 +45,7 @@
 #define WMB asm("wmb")
 #define RMB asm("mb")
 
-static void __inline blas_lock(unsigned long *address){
+static __inline void blas_lock(unsigned long *address){
 #ifndef __DECC
   unsigned long tmp1, tmp2;
   asm volatile(

common_arm.h

@@ -55,7 +55,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #if defined(ARMV6) || defined(ARMV7) || defined(ARMV8)
 
-static void __inline blas_lock(volatile BLASULONG *address){
+static __inline void blas_lock(volatile BLASULONG *address){
 
   int register ret;
 

common_arm64.h

@@ -55,7 +55,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #ifndef ASSEMBLER
 
 
-static void __inline blas_lock(volatile BLASULONG *address){
+static __inline void blas_lock(volatile BLASULONG *address){
 
   BLASULONG ret;
 

common_loongarch64.h

@@ -83,6 +83,19 @@ static inline int blas_quickdivide(blasint x, blasint y){
   return x / y;
 }
 
+#ifndef NO_AFFINITY
+static inline int WhereAmI(void){
+  int ret = 0, counter = 0;
+  __asm__ volatile (
+    "rdtimel.w  %[counter],   %[id]"
+    : [id]"=r"(ret), [counter]"=r"(counter)
+    :
+    : "memory"
+  );
+  return ret;
+}
+#endif
+
 #ifdef DOUBLE
 #define GET_IMAGE(res)  __asm__ __volatile__("fmov.d %0, $f2" : "=f"(res)  : : "memory")
 #else

common_param.h

@@ -1,5 +1,6 @@
 /*********************************************************************/
 /* Copyright 2009, 2010 The University of Texas at Austin.           */
+/* Copyright 2023 The OpenBLAS Project.                              */
 /* All rights reserved.                                              */
 /*                                                                   */
 /* Redistribution and use in source and binary forms, with or        */
@@ -45,12 +46,14 @@
 
 typedef struct {
   int dtb_entries;
+  int switch_ratio;
   int offsetA, offsetB, align;
 
 #if BUILD_BFLOAT16 == 1
   int sbgemm_p, sbgemm_q, sbgemm_r;
   int sbgemm_unroll_m, sbgemm_unroll_n, sbgemm_unroll_mn;
   int sbgemm_align_k;
+  int need_amxtile_permission;  // 0 default, 1 for device support amx.
 
   void   (*sbstobf16_k) (BLASLONG, float    *, BLASLONG, bfloat16 *, BLASLONG);
   void   (*sbdtobf16_k) (BLASLONG, double   *, BLASLONG, bfloat16 *, BLASLONG);

common_power.h

@@ -91,7 +91,7 @@
 
 void *qalloc(int flags, size_t bytes);
 
-static void INLINE blas_lock(volatile unsigned long *address){
+static INLINE void blas_lock(volatile unsigned long *address){
 
   long int ret, val = 1;
 

common_sparc.h

@@ -45,7 +45,7 @@
 
 #ifndef ASSEMBLER
 
-static void __inline blas_lock(volatile unsigned long *address){
+static __inline void blas_lock(volatile unsigned long *address){
 
   long int ret = 1;
 

common_thread.h

@@ -53,7 +53,6 @@ extern void goto_set_num_threads(int nthreads);
 /* Global Parameter */
 extern int blas_cpu_number;
 extern int blas_num_threads;
-extern int blas_num_threads_set;
 extern int blas_omp_linked;
 
 #define BLAS_LEGACY	0x8000U
@@ -136,15 +135,13 @@ typedef struct blas_queue {
 #ifdef SMP_SERVER
 
 extern int blas_server_avail;
+extern int blas_omp_number_max;
 
 static __inline int num_cpu_avail(int level) {
 
 #ifdef USE_OPENMP
 int openmp_nthreads;
-	if (blas_num_threads_set == 0)
 	openmp_nthreads=omp_get_max_threads();
-	else
-	openmp_nthreads=blas_cpu_number;
 #endif
 
 #ifndef USE_OPENMP 
@@ -156,7 +153,13 @@ int openmp_nthreads;
       ) return 1;        
 
 #ifdef USE_OPENMP
-  if (blas_cpu_number != openmp_nthreads) {
+     if (openmp_nthreads > blas_omp_number_max){
+#ifdef DEBUG
+     fprintf(stderr,"WARNING - more OpenMP threads requested (%d) than available (%d)\n",openmp_nthreads,blas_omp_number_max);
+#endif
+     openmp_nthreads = blas_omp_number_max;
+     }
+     if (blas_cpu_number != openmp_nthreads) {
 	  goto_set_num_threads(openmp_nthreads);
   }
 #endif

common_x86.h

@@ -54,7 +54,7 @@
 #define	__volatile__
 #endif
 
-static void __inline blas_lock(volatile BLASULONG *address){
+static __inline void blas_lock(volatile BLASULONG *address){
 
   int ret;
 

common_x86_64.h

@@ -70,7 +70,7 @@
 #define RMB
 #endif
 
-static void __inline blas_lock(volatile BLASULONG *address){
+static __inline void blas_lock(volatile BLASULONG *address){
 
 	
 #ifndef C_MSVC

common_zarch.h

@@ -45,7 +45,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #ifndef ASSEMBLER
 
   /*
-static void __inline blas_lock(volatile BLASULONG *address){
+static __inline void blas_lock(volatile BLASULONG *address){
 
   BLASULONG ret;
 

cpuid_arm64.c

@@ -267,8 +267,9 @@ int detect(void)
 	}
 #else
 #ifdef __APPLE__
-	sysctlbyname("hw.cpufamily",&value,&length,NULL,0);
-	if (value ==131287967|| value == 458787763 ) return CPU_VORTEX;
+	sysctlbyname("hw.cpufamily",&value64,&length64,NULL,0);
+	if (value64 ==131287967|| value64 == 458787763 ) return CPU_VORTEX; //A12/M1
+	if (value64 == 3660830781) return CPU_VORTEX; //A15/M2
 #endif
 	return CPU_ARMV8;	
 #endif

cpuid_loongarch64.c

@@ -32,6 +32,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 **********************************************************************************/
 
 #include <stdint.h>
+#include <sys/auxv.h>
 
 /*  If LASX extension instructions supported,
  *  using core LOONGSON3R5
@@ -46,9 +47,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define CPU_LOONGSON3R5    1
 #define CPU_LOONGSON2K1000 2
 
-#define LOONGARCH_CFG2  0x02
-#define LOONGARCH_LASX  1<<7
-#define LOONGARCH_LSX   1<<6
+#define LA_HWCAP_LSX    (1<<4)
+#define LA_HWCAP_LASX   (1<<5)
 
 static char *cpuname[] = {
   "LOONGSONGENERIC",
@@ -64,17 +64,11 @@ static char *cpuname_lower[] = {
 
 int detect(void) {
 #ifdef __linux
-  uint32_t reg = 0;
+  int flag  = (int)getauxval(AT_HWCAP);
 
-  __asm__ volatile (
-    "cpucfg %0, %1 \n\t"
-    : "+&r"(reg)
-    : "r"(LOONGARCH_CFG2)
-  );
-
-  if (reg & LOONGARCH_LASX)
+  if (flag & LA_HWCAP_LASX)
     return CPU_LOONGSON3R5;
-  else if (reg & LOONGARCH_LSX)
+  else if (flag & LA_HWCAP_LSX)
     return CPU_LOONGSON2K1000;
   else
     return CPU_GENERIC;

cpuid_x86.c

@@ -1479,6 +1479,8 @@ int get_cpuname(void){
           else
 	    return CPUTYPE_NEHALEM;
 	case 15: // Sapphire Rapids
+	  if(support_amx_bf16())
+	    return CPUTYPE_SAPPHIRERAPIDS;
 	  if(support_avx512_bf16())
             return CPUTYPE_COOPERLAKE;	
           if(support_avx512())
@@ -1549,6 +1551,7 @@ int get_cpuname(void){
           case 7: // Raptor Lake
           case 10:
           case 15:
+	  case 14: // Alder Lake N
             if(support_avx2())
               return CPUTYPE_HASWELL;
 	    if(support_avx())
@@ -1845,7 +1848,8 @@ static char *cpuname[] = {
   "ZEN",
   "SKYLAKEX",
   "DHYANA",
-  "COOPERLAKE"
+  "COOPERLAKE",
+  "SAPPHIRERAPIDS",
 };
 
 static char *lowercpuname[] = {
@@ -1902,7 +1906,8 @@ static char *lowercpuname[] = {
   "zen",
   "skylakex",
   "dhyana",
-  "cooperlake"
+  "cooperlake",
+  "sapphirerapids",
 };
 
 static char *corename[] = {
@@ -1936,7 +1941,8 @@ static char *corename[] = {
   "ZEN",
   "SKYLAKEX",
   "DHYANA",
-  "COOPERLAKE"
+  "COOPERLAKE",
+  "SAPPHIRERAPIDS",
 };
 
 static char *corename_lower[] = {
@@ -1970,7 +1976,8 @@ static char *corename_lower[] = {
   "zen",
   "skylakex",
   "dhyana",
-  "cooperlake"
+  "cooperlake",
+  "sapphirerapids",
 };
 
 
@@ -2276,16 +2283,18 @@ int get_coretype(void){
             return CORE_NEHALEM;
 	}
         if (model == 15) { // Sapphire Rapids
+	  if(support_amx_bf16())
+	    return CORE_SAPPHIRERAPIDS;
 	  if(support_avx512_bf16())
-            return CPUTYPE_COOPERLAKE;	
+            return CORE_COOPERLAKE;	
           if(support_avx512())
-            return CPUTYPE_SKYLAKEX;
+            return CORE_SKYLAKEX;
           if(support_avx2())
-            return CPUTYPE_HASWELL;
+            return CORE_HASWELL;
           if(support_avx())
-	    return CPUTYPE_SANDYBRIDGE;
+	    return CORE_SANDYBRIDGE;
 	  else
-	  return CPUTYPE_NEHALEM;	
+	  return CORE_NEHALEM;
         }
       break;
 
@@ -2352,6 +2361,7 @@ int get_coretype(void){
 	  case 7: // Raptor Lake
           case 10:
           case 15:
+	  case 14: // Alder Lake N	  
 #ifndef NO_AVX2
 	  if(support_avx2())
             return CORE_HASWELL;

ctest/Makefile

@@ -208,7 +208,7 @@ FLDFLAGS = $(FFLAGS:-fPIC=) $(LDFLAGS)
 ifeq ($(USE_OPENMP), 1)
 ifeq ($(F_COMPILER), GFORTRAN)
 ifeq ($(C_COMPILER), CLANG)
-CEXTRALIB = -lomp
+CEXTRALIB += -lomp
 endif
 endif
 ifeq ($(F_COMPILER), NAG)

docs/distributing.md

@@ -0,0 +1,270 @@
+# Guidance for redistributing OpenBLAS
+
+*We note that this document contains recommendations only - packagers and other
+redistributors are in charge of how OpenBLAS is built and distributed in their
+systems, and may have good reasons to deviate from the guidance given on this
+page. These recommendations are aimed at general packaging systems, with a user
+base that typically is large, open source (or freely available at least), and
+doesn't behave uniformly or that the packager is directly connected with.*
+
+OpenBLAS has a large number of build-time options which can be used to change
+how it behaves at runtime, how artifacts or symbols are named, etc. Variation
+in build configuration can be necessary to acheive a given end goal within a
+distribution or as an end user. However, such variation can also make it more
+difficult to build on top of OpenBLAS and ship code or other packages in a way
+that works across many different distros. Here we provide guidance about the
+most important build options, what effects they may have when changed, and
+which ones to default to.
+
+The Make and CMake build systems provide equivalent options and yield more or
+less the same artifacts, but not exactly (the CMake builds are still
+experimental). You can choose either one and the options will function in the
+same way, however the CMake outputs may require some renaming. To review
+available build options, see `Makefile.rule` or `CMakeLists.txt` in the root of
+the repository.
+
+Build options typically fall into two categories: (a) options that affect the
+user interface, such as library and symbol names or APIs that are made
+available, and (b) options that affect performance and runtime behavior, such
+as threading behavior or CPU architecture-specific code paths. The user
+interface options are more important to keep aligned between distributions,
+while for the performance-related options there are typically more reasons to
+make choices that deviate from the defaults.
+
+Here are recommendations for user interface related packaging choices where it
+is not likely to be a good idea to deviate (typically these are the default
+settings):
+
+1. Include CBLAS. The CBLAS interface is widely used and it doesn't affect
+   binary size much, so don't turn it off.
+2. Include LAPACK and LAPACKE. The LAPACK interface is also widely used, and
+   while it does make up a significant part of the binary size of the installed
+   library, that does not outweigh the regression in usability when deviating
+   from the default here.[^1]
+3. Always distribute the pkg-config (`.pc`) and CMake `.cmake`) dependency
+   detection files. These files are used by build systems when users want to
+   link against OpenBLAS, and there is no benefit of leaving them out.
+4. Provide the LP64 interface by default, and if in addition to that you choose
+   to provide an ILP64 interface build as well, use a symbol suffix to avoid
+   symbol name clashes (see the next section).
+
+[^1] All major distributions do include LAPACK as of mid 2023 as far as we
+know. Older versions of Arch Linux did not, and that was known to cause
+problems.
+
+
+## ILP64 interface builds
+
+The LP64 (32-bit integer) interface is the default build, and has
+well-established C and Fortran APIs as determined by the reference (Netlib)
+BLAS and LAPACK libraries. The ILP64 (64-bit integer) interface however does
+not have a standard API: symbol names and shared/static library names can be
+produced in multiple ways, and this tends to make it difficult to use.
+As of today there is an agreed-upon way of choosing names for OpenBLAS between
+a number of key users/redistributors, which is the closest thing to a standard
+that there is now. However, there is an ongoing standardization effort in the
+reference BLAS and LAPACK libraries, which differs from the current OpenBLAS
+agreed-upon convention. In this section we'll aim to explain both.
+
+Those two methods are fairly similar, and have a key thing in common: *using a
+symbol suffix*. This is good practice; it is recommended that if you distribute
+an ILP64 build, to have it use a symbol suffix containing `64` in the name.
+This avoids potential symbol clashes when different packages which depend on
+OpenBLAS load both an LP64 and an ILP64 library into memory at the same time.
+
+### The current OpenBLAS agreed-upon ILP64 convention
+
+This convention comprises the shared library name and the symbol suffix in the
+shared library. The symbol suffix to use is `64_`, implying that the library
+name will be `libopenblas64_.so` and the symbols in that library end in `64_`.
+The central issue where this was discussed is
+[openblas#646](https://github.com/xianyi/OpenBLAS/issues/646), and adopters
+include Fedora, Julia, NumPy and SciPy - SuiteSparse already used it as well.
+
+To build shared and static libraries with the currently recommended ILP64
+conventions with Make:
+```bash
+$ make INTERFACE64=1 SYMBOLSUFFIX=64_
+```
+
+This will produce libraries named `libopenblas64_.so|a`, a pkg-config file
+named `openblas64.pc`, and CMake and header files.
+
+Installing locally and inspecting the output will show a few more details:
+```bash
+$ make install PREFIX=$PWD/../openblas/make64 INTERFACE64=1 SYMBOLSUFFIX=64_
+$ tree .  # output slightly edited down
+.
+├── include
+│   ├── cblas.h
+│   ├── f77blas.h
+│   ├── lapacke_config.h
+│   ├── lapacke.h
+│   ├── lapacke_mangling.h
+│   ├── lapacke_utils.h
+│   ├── lapack.h
+│   └── openblas_config.h
+└── lib
+    ├── cmake
+    │   └── openblas
+    │       ├── OpenBLASConfig.cmake
+    │       └── OpenBLASConfigVersion.cmake
+    ├── libopenblas64_.a
+    ├── libopenblas64_.so
+    └── pkgconfig
+        └── openblas64.pc
+```
+
+A key point are the symbol names. These will equal the LP64 symbol names, then
+(for Fortran only) the compiler mangling, and then the `64_` symbol suffix.
+Hence to obtain the final symbol names, we need to take into account which
+Fortran compiler we are using. For the most common cases (e.g., gfortran, Intel
+Fortran, or Flang), that means appending a single underscore. In that case, the
+result is:
+
+| base API name | binary symbol name | call from Fortran code | call from C code      |
+|---------------|--------------------|------------------------|-----------------------|
+| `dgemm`       | `dgemm_64_`        | `dgemm_64(...)`        | `dgemm_64_(...)`      |
+| `cblas_dgemm` | `cblas_dgemm64_`   | n/a                    | `cblas_dgemm64_(...)` |
+
+It is quite useful to have these symbol names be as uniform as possible across
+different packaging systems.
+
+The equivalent build options with CMake are:
+```bash
+$ mkdir build && cd build
+$ cmake .. -DINTERFACE64=1 -DSYMBOLSUFFIX=64_ -DBUILD_SHARED_LIBS=ON -DBUILD_STATIC_LIBS=ON
+$ cmake --build . -j
+```
+
+Note that the result is not 100% identical to the Make result. For example, the
+library name ends in `_64` rather than `64_` - it is recommended to rename them
+to match the Make library names (also update the `libsuffix` entry in
+`openblas64.pc` to match that rename).
+```bash
+$ cmake --install . --prefix $PWD/../../openblas/cmake64
+$ tree .
+.
+├── include
+│   └── openblas64
+│       ├── cblas.h
+│       ├── f77blas.h
+│       ├── lapacke_config.h
+│       ├── lapacke_example_aux.h
+│       ├── lapacke.h
+│       ├── lapacke_mangling.h
+│       ├── lapacke_utils.h
+│       ├── lapack.h
+│       ├── openblas64
+│       │   └── lapacke_mangling.h
+│       └── openblas_config.h
+└── lib
+    ├── cmake
+    │   └── OpenBLAS64
+    │       ├── OpenBLAS64Config.cmake
+    │       ├── OpenBLAS64ConfigVersion.cmake
+    │       ├── OpenBLAS64Targets.cmake
+    │       └── OpenBLAS64Targets-noconfig.cmake
+    ├── libopenblas_64.a
+    ├── libopenblas_64.so -> libopenblas_64.so.0
+    └── pkgconfig
+        └── openblas64.pc
+```
+
+
+### The upcoming standardized ILP64 convention
+
+While the `64_` convention above got some adoption, it's slightly hacky and is
+implemented through the use of `objcopy`. An effort is ongoing for a more
+broadly adopted convention in the reference BLAS and LAPACK libraries, using
+(a) the `_64` suffix, and (b) applying that suffix _before_ rather than after
+Fortran compiler mangling. The central issue for this is
+[lapack#666](https://github.com/Reference-LAPACK/lapack/issues/666).
+
+For the most common cases of compiler mangling (a single `_` appended), the end
+result will be:
+
+| base API name | binary symbol name | call from Fortran code | call from C code      |
+|---------------|--------------------|------------------------|-----------------------|
+| `dgemm`       | `dgemm_64_`        | `dgemm_64(...)`        | `dgemm_64_(...)`      |
+| `cblas_dgemm` | `cblas_dgemm_64`   | n/a                    | `cblas_dgemm_64(...)` |
+
+For other compiler mangling schemes, replace the trailing `_` by the scheme in use.
+
+The shared library name for this `_64` convention should be `libopenblas_64.so`.
+
+Note: it is not yet possible to produce an OpenBLAS build which employs this
+convention! Once reference BLAS and LAPACK with support for `_64` have been
+released, a future OpenBLAS release will support it. For now, please use the
+older `64_` scheme and avoid using the name `libopenblas_64.so`; it should be
+considered reserved for future use of the `_64` standard as prescribed by
+reference BLAS/LAPACK.
+
+
+## Performance and runtime behavior related build options
+
+For these options there are multiple reasonable or common choices.
+
+### Threading related options
+
+OpenBLAS can be built as a multi-threaded or single-threaded library, with the
+default being multi-threaded. It's expected that the default `libopenblas`
+library is multi-threaded; if you'd like to also distribute single-threaded
+builds, consider naming them `libopenblas_sequential`.
+
+OpenBLAS can be built with pthreads or OpenMP as the threading model, with the
+default being pthreads. Both options are commonly used, and the choice here
+should not influence the shared library name. The choice will be captured by
+the `.pc` file. E.g.,:
+```bash
+$ pkg-config --libs openblas
+-fopenmp -lopenblas
+
+$ cat openblas.pc
+...
+openblas_config= ... USE_OPENMP=0 MAX_THREADS=24
+```
+
+The maximum number of threads users will be able to use is determined at build
+time by the `NUM_THREADS` build option. It defaults to 24, and there's a wide
+range of values that are reasonable to use (up to 256). 64 is a typical choice
+here; there is a memory footprint penalty that is linear in `NUM_THREADS`.
+Please see `Makefile.rule` for more details.
+
+### CPU architecture related options
+
+OpenBLAS contains a lot of CPU architecture-specific optimizations, hence when
+distributing to a user base with a variety of hardware, it is recommended to
+enable CPU architecture runtime detection. This will dynamically select
+optimized kernels for individual APIs. To do this, use the `DYNAMIC_ARCH=1`
+build option. This is usually done on all common CPU families, except when
+there are known issues.
+
+In case the CPU architecture is known (e.g. you're building binaries for macOS
+M1 users), it is possible to specify the target architecture directly with the
+`TARGET=` build option.
+
+`DYNAMIC_ARCH` and `TARGET` are covered in more detail in the main `README.md`
+in this repository.
+
+
+## Real-world examples
+
+OpenBLAS is likely to be distributed in one of these distribution models:
+
+1. As a standalone package, or multiple packages, in a packaging ecosystem like
+   a Linux distro, Homebrew, conda-forge or MSYS2.
+2. Vendored as part of a larger package, e.g. in Julia, NumPy, SciPy, or R.
+3. Locally, e.g. making available as a build on a single HPC cluster.
+
+The guidance on this page is most important for models (1) and (2). These links
+to build recipes for a representative selection of packaging systems may be
+helpful as a reference:
+
+- [Fedora](https://src.fedoraproject.org/rpms/openblas/blob/rawhide/f/openblas.spec)
+- [Debian](https://salsa.debian.org/science-team/openblas/-/blob/master/debian/rules)
+- [Homebrew](https://github.com/Homebrew/homebrew-core/blob/HEAD/Formula/openblas.rb)
+- [MSYS2](https://github.com/msys2/MINGW-packages/blob/master/mingw-w64-openblas/PKGBUILD)
+- [conda-forge](https://github.com/conda-forge/openblas-feedstock/blob/main/recipe/build.sh)
+- [NumPy/SciPy](https://github.com/MacPython/openblas-libs/blob/main/tools/build_openblas.sh)
+- [Nixpkgs](https://github.com/NixOS/nixpkgs/blob/master/pkgs/development/libraries/science/math/openblas/default.nix)

driver/level3/level3_gemm3m_thread.c

@@ -1,5 +1,6 @@
 /*********************************************************************/
 /* Copyright 2009, 2010 The University of Texas at Austin.           */
+/* Copyright 2023 The OpenBLAS Project.                              */
 /* All rights reserved.                                              */
 /*                                                                   */
 /* Redistribution and use in source and binary forms, with or        */
@@ -44,10 +45,6 @@
 #define DIVIDE_RATE 2
 #endif
 
-#ifndef SWITCH_RATIO
-#define SWITCH_RATIO 2
-#endif
-
 //The array of job_t may overflow the stack.
 //Instead, use malloc to alloc job_t.
 #if MAX_CPU_NUMBER > BLAS3_MEM_ALLOC_THRESHOLD
@@ -1015,6 +1012,12 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
   BLASLONG divN, divT;
   int mode;
 
+#if defined(DYNAMIC_ARCH)
+  int switch_ratio = gotoblas->switch_ratio;
+#else
+  int switch_ratio = SWITCH_RATIO;
+#endif
+
   if (range_m) {
     BLASLONG m_from = *(((BLASLONG *)range_m) + 0);
     BLASLONG m_to   = *(((BLASLONG *)range_m) + 1);
@@ -1030,7 +1033,7 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
   }
 */
 
-  if ((args -> m < nthreads * SWITCH_RATIO) || (args -> n < nthreads * SWITCH_RATIO)) {
+  if ((args -> m < nthreads * switch_ratio) || (args -> n < nthreads * switch_ratio)) {
     GEMM3M_LOCAL(args, range_m, range_n, sa, sb, 0);
     return 0;
   }
@@ -1038,7 +1041,7 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
   divT = nthreads;
   divN = 1;
 
-  while ((GEMM3M_P * divT > m * SWITCH_RATIO) && (divT > 1)) {
+  while ((GEMM3M_P * divT > m * switch_ratio) && (divT > 1)) {
     do {
       divT --;
       divN = 1;

driver/level3/level3_syrk_threaded.c

@@ -1,5 +1,6 @@
 /*********************************************************************/
 /* Copyright 2009, 2010 The University of Texas at Austin.           */
+/* Copyright 2023 The OpenBLAS Project.                              */
 /* All rights reserved.                                              */
 /*                                                                   */
 /* Redistribution and use in source and binary forms, with or        */
@@ -44,10 +45,6 @@
 #define DIVIDE_RATE 2
 #endif
 
-#ifndef SWITCH_RATIO
-#define SWITCH_RATIO 2
-#endif
-
 //The array of job_t may overflow the stack.
 //Instead, use malloc to alloc job_t.
 #if MAX_CPU_NUMBER > BLAS3_MEM_ALLOC_THRESHOLD
@@ -528,7 +525,13 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
   int  mode, mask;
   double dnum, di, dinum;
 
-  if ((nthreads  == 1) || (args -> n < nthreads * SWITCH_RATIO)) {
+#if defined(DYNAMIC_ARCH)
+  int switch_ratio = gotoblas->switch_ratio;
+#else
+  int switch_ratio = SWITCH_RATIO;
+#endif
+
+  if ((nthreads == 1) || (args->n < nthreads * switch_ratio)) {
     SYRK_LOCAL(args, range_m, range_n, sa, sb, 0);
     return 0;
   }

driver/level3/level3_thread.c

@@ -1,5 +1,6 @@
 /*********************************************************************/
 /* Copyright 2009, 2010 The University of Texas at Austin.           */
+/* Copyright 2023 The OpenBLAS Project.                              */
 /* All rights reserved.                                              */
 /*                                                                   */
 /* Redistribution and use in source and binary forms, with or        */
@@ -44,10 +45,6 @@
 #define DIVIDE_RATE 2
 #endif
 
-#ifndef SWITCH_RATIO
-#define SWITCH_RATIO 2
-#endif
-
 #ifndef GEMM_PREFERED_SIZE
 #define GEMM_PREFERED_SIZE 1
 #endif
@@ -577,6 +574,11 @@ InitializeCriticalSection((PCRITICAL_SECTION)&level3_lock);
   BLASLONG width, i, j, k, js;
   BLASLONG m, n, n_from, n_to;
   int mode;
+#if defined(DYNAMIC_ARCH)
+  int switch_ratio = gotoblas->switch_ratio;
+#else
+  int switch_ratio = SWITCH_RATIO;
+#endif
 
   /* Get execution mode */
 #ifndef COMPLEX
@@ -698,8 +700,8 @@ EnterCriticalSection((PCRITICAL_SECTION)&level3_lock);
     num_parts  = 0;
     while (n > 0){
       width = blas_quickdivide(n + nthreads - num_parts - 1, nthreads - num_parts);
-      if (width < SWITCH_RATIO) {
-        width = SWITCH_RATIO;
+      if (width < switch_ratio) {
+        width = switch_ratio;
       }
       width = round_up(n, width, GEMM_PREFERED_SIZE);
 
@@ -746,6 +748,11 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, IFLOAT *sa, IF
   BLASLONG m = args -> m;
   BLASLONG n = args -> n;
   BLASLONG nthreads_m, nthreads_n;
+#if defined(DYNAMIC_ARCH)
+  int switch_ratio = gotoblas->switch_ratio;
+#else
+  int switch_ratio = SWITCH_RATIO;
+#endif
 
   /* Get dimensions from index ranges if available */
   if (range_m) {
@@ -755,21 +762,21 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, IFLOAT *sa, IF
     n = range_n[1] - range_n[0];
   }
 
-  /* Partitions in m should have at least SWITCH_RATIO rows */
-  if (m < 2 * SWITCH_RATIO) {
+  /* Partitions in m should have at least switch_ratio rows */
+  if (m < 2 * switch_ratio) {
     nthreads_m = 1;
   } else {
     nthreads_m = args -> nthreads;
-    while (m < nthreads_m * SWITCH_RATIO) {
+    while (m < nthreads_m * switch_ratio) {
       nthreads_m = nthreads_m / 2;
     }
   }
 
-  /* Partitions in n should have at most SWITCH_RATIO * nthreads_m columns */
-  if (n < SWITCH_RATIO * nthreads_m) {
+  /* Partitions in n should have at most switch_ratio * nthreads_m columns */
+  if (n < switch_ratio * nthreads_m) {
     nthreads_n = 1;
   } else {
-    nthreads_n = (n + SWITCH_RATIO * nthreads_m - 1) / (SWITCH_RATIO * nthreads_m);
+    nthreads_n = (n + switch_ratio * nthreads_m - 1) / (switch_ratio * nthreads_m);
     if (nthreads_m * nthreads_n > args -> nthreads) {
       nthreads_n = blas_quickdivide(args -> nthreads, nthreads_m);
     }

driver/others/blas_server.c

@@ -973,7 +973,7 @@ void goto_set_num_threads(int num_threads) {
 
     increased_threads = 1;
 
-    for(i = blas_num_threads - 1; i < num_threads - 1; i++){
+    for(i = (blas_num_threads > 0) ? blas_num_threads - 1 : 0; i < num_threads - 1; i++){
 
       atomic_store_queue(&thread_status[i].queue, (blas_queue_t *)0);
       thread_status[i].status = THREAD_STATUS_WAKEUP;

driver/others/blas_server_omp.c

@@ -68,6 +68,7 @@
 #endif
 
 int blas_server_avail = 0;
+int blas_omp_number_max = 0;
 
 extern int openblas_omp_adaptive_env();
 
@@ -100,8 +101,6 @@ static void adjust_thread_buffers() {
 
 void goto_set_num_threads(int num_threads) {
 
-  blas_num_threads_set = 1;
-  if (num_threads < 0) blas_num_threads_set = 0;
   if (num_threads < 1) num_threads = blas_num_threads;
 
   if (num_threads > MAX_CPU_NUMBER) num_threads = MAX_CPU_NUMBER;
@@ -125,6 +124,8 @@ void openblas_set_num_threads(int num_threads) {
 }
 
 int blas_thread_init(void){
+if(blas_omp_number_max <= 0)
+  blas_omp_number_max = omp_get_max_threads();
 
   blas_get_cpu_number();
 

driver/others/blas_server_win32.c

@@ -568,7 +568,7 @@ void goto_set_num_threads(int num_threads)
 			blas_server_avail = 1;
 		}
 
-		for(i = blas_num_threads - 1; i < num_threads - 1; i++){
+		for(i = (blas_num_threads > 0) ? blas_num_threads - 1 : 0; i < num_threads - 1; i++){
 
 			blas_threads[i] = CreateThread(NULL, 0,
 				     blas_thread_server, (void *)i,

driver/others/dynamic.c

@@ -220,6 +220,19 @@ extern gotoblas_t gotoblas_COOPERLAKE;
 #else
 #define gotoblas_COOPERLAKE gotoblas_PRESCOTT
 #endif
+#ifdef DYN_SAPPHIRERAPIDS
+extern gotoblas_t gotoblas_SAPPHIRERAPIDS;
+#elif defined(DYN_SKYLAKEX)
+#define gotoblas_SAPPHIRERAPIDS gotoblas_SKYLAKEX
+#elif defined(DYN_HASWELL)
+#define gotoblas_SAPPHIRERAPIDS gotoblas_HASWELL
+#elif defined(DYN_SANDYBRIDGE)
+#define gotoblas_SAPPHIRERAPIDS gotoblas_SANDYBRIDGE
+#elif defined(DYN_NEHALEM)
+#define gotoblas_SAPPHIRERAPIDS gotoblas_NEHALEM
+#else
+#define gotoblas_SAPPHIRERAPIDS gotoblas_PRESCOTT
+#endif
 
 
 #else // not DYNAMIC_LIST
@@ -268,9 +281,11 @@ extern gotoblas_t  gotoblas_ZEN;
 #ifndef NO_AVX512
 extern gotoblas_t  gotoblas_SKYLAKEX;
 extern gotoblas_t  gotoblas_COOPERLAKE;
+extern gotoblas_t  gotoblas_SAPPHIRERAPIDS;
 #else
 #define gotoblas_SKYLAKEX gotoblas_HASWELL
 #define gotoblas_COOPERLAKE gotoblas_HASWELL
+#define gotoblas_SAPPHIRERAPIDS gotoblas_HASWELL
 #endif
 #endif
 #else
@@ -279,6 +294,7 @@ extern gotoblas_t  gotoblas_COOPERLAKE;
 #define gotoblas_HASWELL gotoblas_NEHALEM
 #define gotoblas_SKYLAKEX gotoblas_NEHALEM
 #define gotoblas_COOPERLAKE gotoblas_NEHALEM
+#define gotoblas_SAPPHIRERAPIDS gotoblas_NEHALEM
 #define gotoblas_BULLDOZER gotoblas_BARCELONA
 #define gotoblas_PILEDRIVER gotoblas_BARCELONA
 #define gotoblas_STEAMROLLER gotoblas_BARCELONA
@@ -378,6 +394,31 @@ int support_avx512_bf16(){
 #endif
 }
 
+#define BIT_AMX_TILE	0x01000000
+#define BIT_AMX_BF16	0x00400000
+#define BIT_AMX_ENBD	0x00060000
+
+int support_amx_bf16() {
+#if !defined(NO_AVX) && !defined(NO_AVX512)
+  int eax, ebx, ecx, edx;
+  int ret=0;
+
+  if (!support_avx512())
+    return 0;
+  // CPUID.7.0:EDX indicates AMX support
+  cpuid_count(7, 0, &eax, &ebx, &ecx, &edx);
+  if ((edx & BIT_AMX_TILE) && (edx & BIT_AMX_BF16)) {
+    // CPUID.D.0:EAX[17:18] indicates AMX enabled
+    cpuid_count(0xd, 0, &eax, &ebx, &ecx, &edx);
+    if ((eax & BIT_AMX_ENBD) == BIT_AMX_ENBD)
+      ret = 1;
+  }
+  return ret;
+#else
+  return 0;
+#endif
+}
+
 extern void openblas_warning(int verbose, const char * msg);
 #define FALLBACK_VERBOSE 1
 #define NEHALEM_FALLBACK "OpenBLAS : Your OS does not support AVX instructions. OpenBLAS is using Nehalem kernels as a fallback, which may give poorer performance.\n"
@@ -689,6 +730,8 @@ static gotoblas_t *get_coretype(void){
 	  }
 	}
 	if (model == 15){          // Sapphire Rapids
+	   if(support_amx_bf16())
+	     return &gotoblas_SAPPHIRERAPIDS;
 	   if(support_avx512_bf16())
              return &gotoblas_COOPERLAKE;
           if (support_avx512()) 
@@ -941,7 +984,8 @@ static char *corename[] = {
     "Excavator",
     "Zen",
     "SkylakeX",
-    "Cooperlake"
+    "Cooperlake",
+    "SapphireRapids"
 };
 
 char *gotoblas_corename(void) {
@@ -1006,6 +1050,7 @@ char *gotoblas_corename(void) {
   if (gotoblas == &gotoblas_ZEN)          return corename[23];
   if (gotoblas == &gotoblas_SKYLAKEX)     return corename[24];
   if (gotoblas == &gotoblas_COOPERLAKE)   return corename[25];
+  if (gotoblas == &gotoblas_SAPPHIRERAPIDS) return corename[26];
   return corename[0];
 }
 

driver/others/dynamic_arm64.c

@@ -1,5 +1,6 @@
 /*********************************************************************/
 /* Copyright 2009, 2010 The University of Texas at Austin.           */
+/* Copyright 2023 The OpenBLAS Project                               */
 /* All rights reserved.                                              */
 /*                                                                   */
 /* Redistribution and use in source and binary forms, with or        */
@@ -109,6 +110,11 @@ extern gotoblas_t gotoblas_NEOVERSEN2;
 #else
 #define gotoblas_NEOVERSEN2 gotoblas_ARMV8
 #endif
+#ifdef DYN_ARMV8SVE
+extern gotoblas_t gotoblas_ARMV8SVE;
+#else
+#define gotoblas_ARMV8SVE gotoblas_ARMV8
+#endif
 #ifdef DYN_CORTEX_A55
 extern gotoblas_t  gotoblas_CORTEXA55;
 #else
@@ -128,17 +134,21 @@ extern gotoblas_t  gotoblas_NEOVERSEN1;
 #ifndef NO_SVE
 extern gotoblas_t  gotoblas_NEOVERSEV1;
 extern gotoblas_t  gotoblas_NEOVERSEN2;
+extern gotoblas_t  gotoblas_ARMV8SVE;
 #else
 #define gotoblas_NEOVERSEV1 gotoblas_ARMV8
 #define gotoblas_NEOVERSEN2 gotoblas_ARMV8
+#define gotoblas_ARMV8SVE   gotoblas_ARMV8
 #endif
 extern gotoblas_t  gotoblas_THUNDERX3T110;
 extern gotoblas_t  gotoblas_CORTEXA55;
 #endif
 
 extern void openblas_warning(int verbose, const char * msg);
+#define FALLBACK_VERBOSE 1
+#define NEOVERSEN1_FALLBACK "OpenBLAS : Your OS does not support SVE instructions. OpenBLAS is using Neoverse N1 kernels as a fallback, which may give poorer performance.\n"
 
-#define NUM_CORETYPES   13
+#define NUM_CORETYPES   16
 
 /*
  * In case asm/hwcap.h is outdated on the build system, make sure
@@ -147,6 +157,9 @@ extern void openblas_warning(int verbose, const char * msg);
 #ifndef HWCAP_CPUID
 #define HWCAP_CPUID (1 << 11)
 #endif
+#ifndef HWCAP_SVE
+#define HWCAP_SVE (1 << 22)
+#endif
 
 #define get_cpu_ftr(id, var) ({					\
 		__asm__ __volatile__ ("mrs %0, "#id : "=r" (var));		\
@@ -168,6 +181,7 @@ static char *corename[] = {
   "neoversen2",
   "thunderx3t110",
   "cortexa55",
+  "armv8sve",
   "unknown"
 };
 
@@ -187,6 +201,7 @@ char *gotoblas_corename(void) {
   if (gotoblas == &gotoblas_NEOVERSEN2)   return corename[12];
   if (gotoblas == &gotoblas_THUNDERX3T110) return corename[13];
   if (gotoblas == &gotoblas_CORTEXA55)    return corename[14];
+  if (gotoblas == &gotoblas_ARMV8SVE)     return corename[15];
   return corename[NUM_CORETYPES];
 }
 
@@ -221,6 +236,7 @@ static gotoblas_t *force_coretype(char *coretype) {
     case 12: return (&gotoblas_NEOVERSEN2);
     case 13: return (&gotoblas_THUNDERX3T110);
     case 14: return (&gotoblas_CORTEXA55);
+    case 15: return (&gotoblas_ARMV8SVE);
   }
   snprintf(message, 128, "Core not found: %s\n", coretype);
   openblas_warning(1, message);
@@ -281,9 +297,17 @@ static gotoblas_t *get_coretype(void) {
           return &gotoblas_NEOVERSEN1;
 #ifndef NO_SVE
         case 0xd49:
-          return &gotoblas_NEOVERSEN2;
+	  if (!(getauxval(AT_HWCAP) & HWCAP_SVE)) {
+	    openblas_warning(FALLBACK_VERBOSE, NEOVERSEN1_FALLBACK);  
+	    return &gotoblas_NEOVERSEN1;
+	  } else
+            return &gotoblas_NEOVERSEN2;
 	case 0xd40:
-	  return &gotoblas_NEOVERSEV1;
+	  if (!(getauxval(AT_HWCAP) & HWCAP_SVE)) {
+		  openblas_warning(FALLBACK_VERBOSE, NEOVERSEN1_FALLBACK);
+	    return &gotoblas_NEOVERSEN1;
+      	  }else
+	    return &gotoblas_NEOVERSEV1;
 #endif
 	case 0xd05: // Cortex A55
 	  return &gotoblas_CORTEXA55;
@@ -332,6 +356,12 @@ static gotoblas_t *get_coretype(void) {
       snprintf(coremsg, 128, "Unknown CPU model - implementer %x part %x\n",implementer,part);
       openblas_warning(1, coremsg);
   }
+#ifndef NO_SVE
+  if ((getauxval(AT_HWCAP) & HWCAP_SVE)) {
+    return &gotoblas_ARMV8SVE;
+  }
+#endif
+
   return NULL;
 #endif
 }

driver/others/memory.c

@@ -422,8 +422,6 @@ This value is equal or large than blas_cpu_number. This means some threads are s
 */
 int blas_num_threads = 0;
 
-int blas_num_threads_set = 0;
-
 int  goto_get_num_procs  (void) {
   return blas_cpu_number;
 }
@@ -1996,8 +1994,6 @@ This value is equal or large than blas_cpu_number. This means some threads are s
 */
 int blas_num_threads = 0;
 
-int blas_num_threads_set = 0;
-
 int  goto_get_num_procs  (void) {
   return blas_cpu_number;
 }
@@ -3015,6 +3011,8 @@ void *blas_memory_alloc(int procpos){
 #endif
  if (memory_overflowed) goto terminate;
   fprintf(stderr,"OpenBLAS warning: precompiled NUM_THREADS exceeded, adding auxiliary array for thread metadata.\n");
+  fprintf(stderr,"To avoid this warning, please rebuild your copy of OpenBLAS with a larger NUM_THREADS setting\n");
+  fprintf(stderr,"or set the environment variable OPENBLAS_NUM_THREADS to %d or lower\n", NUM_BUFFERS);
   memory_overflowed=1;
   new_release_info = (struct release_t*) malloc(512*sizeof(struct release_t));
   newmemory = (struct newmemstruct*) malloc(512*sizeof(struct newmemstruct));

driver/others/memory_qalloc.c

@@ -283,7 +283,6 @@ The numbers of threads in the thread pool.
 This value is equal or large than blas_cpu_number. This means some threads are sleep.
 */
 int blas_num_threads = 0;
-int blas_num_threads_set = 0;
 
 int  goto_get_num_procs  (void) {
   return blas_cpu_number;

exports/gensymbol

@@ -21,7 +21,7 @@ blasobjsc="
     chbmv chemm chemv cher2 cher2k cher cherk scabs1 scamax
     chpmv chpr2 chpr crotg cscal csrot csscal cswap scamin scasum scnrm2
     csymm csyr2k csyrk ctbmv ctbsv ctpmv ctpsv ctrmm ctrmv ctrsm
-    ctrsv icamax icamin cimatcopy comatcopy cgeadd scsum"
+    ctrsv icamax icamin cimatcopy comatcopy cgeadd scsum cgemmt"
 
 blasobjsd="
     damax damin dasum daxpy daxpby dcabs1 dcopy ddot dgbmv dgemm
@@ -29,7 +29,7 @@ blasobjsd="
     dscal dsdot dspmv dspr2 dimatcopy domatcopy
     dspr dswap dsymm dsymv dsyr2 dsyr2k dsyr dsyrk dtbmv dtbsv
     dtpmv dtpsv dtrmm dtrmv dtrsm dtrsv
-        idamax idamin idmax idmin dgeadd dsum"
+        idamax idamin idmax idmin dgeadd dsum dgemmt"
 
 blasobjss="
     isamax isamin ismax ismin
@@ -38,7 +38,7 @@ blasobjss="
     smax smin snrm2 simatcopy somatcopy
     srot srotg srotm srotmg ssbmv sscal sspmv sspr2 sspr sswap
     ssymm ssymv ssyr2 ssyr2k ssyr ssyrk stbmv stbsv stpmv stpsv
-    strmm strmv strsm strsv  sgeadd ssum"
+    strmm strmv strsm strsv  sgeadd ssum sgemmt"
 
 blasobjsz="
     izamax izamin
@@ -48,7 +48,7 @@ blasobjsz="
     zhpr zrotg zscal zswap zsymm zsyr2k zsyrk ztbmv
     ztbsv ztpmv ztpsv ztrmm ztrmv ztrsm ztrsv
     zomatcopy  zimatcopy dzamax dzamin dzasum dznrm2
-    zgeadd  dzsum"
+    zgeadd  dzsum zgemmt"
 
 blasobjs="lsame xerbla"
 bfblasobjs="sbgemm sbgemv sbdot sbstobf16 sbdtobf16 sbf16tos dbf16tod"
@@ -58,7 +58,7 @@ cblasobjsc="
     cblas_cher cblas_cherk  cblas_chpmv cblas_chpr2 cblas_chpr cblas_cscal cblas_caxpby
     cblas_csscal cblas_cswap cblas_csymm cblas_csyr2k cblas_csyrk cblas_ctbmv cblas_cgeadd
     cblas_ctbsv cblas_ctpmv cblas_ctpsv cblas_ctrmm cblas_ctrmv cblas_ctrsm cblas_ctrsv
-    cblas_scnrm2 cblas_scasum
+    cblas_scnrm2 cblas_scasum cblas_cgemmt
     cblas_icamax cblas_icamin cblas_icmin cblas_icmax cblas_scsum cblas_cimatcopy cblas_comatcopy
     "
 cblasobjsd="
@@ -67,7 +67,7 @@ cblasobjsd="
     cblas_drot cblas_drotg cblas_drotm cblas_drotmg cblas_dsbmv cblas_dscal cblas_dsdot
     cblas_dspmv cblas_dspr2 cblas_dspr cblas_dswap cblas_dsymm cblas_dsymv cblas_dsyr2
     cblas_dsyr2k cblas_dsyr cblas_dsyrk cblas_dtbmv cblas_dtbsv cblas_dtpmv cblas_dtpsv
-    cblas_dtrmm cblas_dtrmv cblas_dtrsm cblas_dtrsv cblas_daxpby cblas_dgeadd
+    cblas_dtrmm cblas_dtrmv cblas_dtrsm cblas_dtrsv cblas_daxpby cblas_dgeadd cblas_dgemmt
     cblas_idamax cblas_idamin cblas_idmin cblas_idmax cblas_dsum cblas_dimatcopy cblas_domatcopy
     "
 
@@ -78,7 +78,7 @@ cblasobjss="
     cblas_srotm cblas_srotmg cblas_ssbmv cblas_sscal cblas_sspmv cblas_sspr2 cblas_sspr
     cblas_sswap cblas_ssymm cblas_ssymv cblas_ssyr2 cblas_ssyr2k cblas_ssyr cblas_ssyrk
     cblas_stbmv cblas_stbsv cblas_stpmv cblas_stpsv cblas_strmm cblas_strmv cblas_strsm
-    cblas_strsv cblas_sgeadd
+    cblas_strsv cblas_sgeadd cblas_sgemmt
     cblas_isamax cblas_isamin cblas_ismin cblas_ismax cblas_ssum cblas_simatcopy cblas_somatcopy
     "
 
@@ -89,7 +89,7 @@ cblasobjsz="
     cblas_zhpr cblas_zscal cblas_zswap cblas_zsymm cblas_zsyr2k cblas_zsyrk
     cblas_ztbmv cblas_ztbsv cblas_ztpmv cblas_ztpsv cblas_ztrmm cblas_ztrmv cblas_ztrsm
     cblas_ztrsv cblas_cdotc_sub cblas_cdotu_sub cblas_zdotc_sub cblas_zdotu_sub
-    cblas_zaxpby cblas_zgeadd
+    cblas_zaxpby cblas_zgeadd cblas_zgemmt
     cblas_izamax cblas_izamin cblas_izmin cblas_izmax cblas_dzsum cblas_zimatcopy cblas_zomatcopy
 "
 
@@ -716,6 +716,7 @@ lapackobjs2z="$lapackobjs2z
 # functions added for lapack-3.7.0
 lapackobjs2s="$lapackobjs2s
     slarfy
+    ssyconvf
     strevc3
     sgelqt
     sgelqt3
@@ -843,6 +844,23 @@ lapackobjs2z="$lapackobjs2z
     zungtsqr_row
     "
 
+#functions added for lapack-3.11
+lapackobjs2c="$lapackobjs2c
+    cgedmd
+    cgedmdq
+    "
+lapackobjs2d="$lapackobjs2d
+    dgedmd
+    dgedmdq
+    "
+lapackobjs2s="$lapackobjs2s
+    sgedmd
+    sgedmdq
+    "
+lapackobjs2z="$lapackobjs2z
+    zgedmd
+    zgedmdq
+    "
 lapack_extendedprecision_objs="
     zposvxx clagge clatms chesvxx cposvxx cgesvxx ssyrfssx csyrfsx
     dlagsy dsysvxx sporfsx slatms zlatms zherfsx csysvxx
@@ -1012,6 +1030,10 @@ lapackeobjsc="
     LAPACKE_cgebrd_work
     LAPACKE_cgecon
     LAPACKE_cgecon_work
+    LAPACKE_cgedmd
+    LAPACKE_cgedmd_work
+    LAPACKE_cgedmdq
+    LAPACKE_cgedmdq_work
     LAPACKE_cgeequ
     LAPACKE_cgeequ_work
     LAPACKE_cgeequb
@@ -1671,6 +1693,10 @@ lapackeobjsd="
     LAPACKE_dgebrd_work
     LAPACKE_dgecon
     LAPACKE_dgecon_work
+    LAPACKE_dgedmd
+    LAPACKE_dgedmd_work
+    LAPACKE_dgedmdq
+    LAPACKE_dgedmdq_work
     LAPACKE_dgeequ
     LAPACKE_dgeequ_work
     LAPACKE_dgeequb
@@ -2284,6 +2310,10 @@ lapackeobjss="
     LAPACKE_sgebrd_work
     LAPACKE_sgecon
     LAPACKE_sgecon_work
+    LAPACKE_sgedmd
+    LAPACKE_sgedmd_work
+    LAPACKE_sgedmdq
+    LAPACKE_sgedmdq_work
     LAPACKE_sgeequ
     LAPACKE_sgeequ_work
     LAPACKE_sgeequb
@@ -2893,6 +2923,10 @@ lapackeobjsz="
     LAPACKE_zgebrd_work
     LAPACKE_zgecon
     LAPACKE_zgecon_work
+    LAPACKE_zgedmd
+    LAPACKE_zgedmd_work
+    LAPACKE_zgedmdq
+    LAPACKE_zgedmdq_work
     LAPACKE_zgeequ
     LAPACKE_zgeequ_work
     LAPACKE_zgeequb

exports/gensymbol.pl

@@ -21,7 +21,7 @@
     chbmv,chemm,chemv,cher2,cher2k,cher,cherk,scabs1,scamax,
     chpmv,chpr2,chpr,crotg,cscal,csrot,csscal,cswap,scamin,scasum,scnrm2,
     csymm,csyr2k,csyrk,ctbmv,ctbsv,ctpmv,ctpsv,ctrmm,ctrmv,ctrsm,
-    ctrsv,icamax,icamin,cimatcopy,comatcopy,cgeadd,scsum);
+    ctrsv,icamax,icamin,cimatcopy,comatcopy,cgeadd,scsum,cgemmt);
     
 @blasobjsd = (
     damax,damin,dasum,daxpy,daxpby,dcabs1,dcopy,ddot,dgbmv,dgemm,
@@ -29,7 +29,7 @@
     dscal,dsdot,dspmv,dspr2,dimatcopy,domatcopy,
     dspr,dswap,dsymm,dsymv,dsyr2,dsyr2k,dsyr,dsyrk,dtbmv,dtbsv,
     dtpmv,dtpsv,dtrmm,dtrmv,dtrsm,dtrsv,
-        idamax,idamin,idmax,idmin,dgeadd,dsum);
+        idamax,idamin,idmax,idmin,dgeadd,dsum,dgemmt);
     
 @blasobjss = (
     isamax,isamin,ismax,ismin,
@@ -38,7 +38,7 @@
     smax,smin,snrm2,simatcopy,somatcopy,
     srot,srotg,srotm,srotmg,ssbmv,sscal,sspmv,sspr2,sspr,sswap,
     ssymm,ssymv,ssyr2,ssyr2k,ssyr,ssyrk,stbmv,stbsv,stpmv,stpsv,
-    strmm,strmv,strsm,strsv, sgeadd,ssum);
+    strmm,strmv,strsm,strsv, sgeadd,ssum,sgemmt);
      
 @blasobjsz = (
     izamax,izamin,,
@@ -48,7 +48,7 @@
     zhpr,zrotg,zscal,zswap,zsymm,zsyr2k,zsyrk,ztbmv,
     ztbsv,ztpmv,ztpsv,ztrmm,ztrmv,ztrsm,ztrsv,
     zomatcopy, zimatcopy,dzamax,dzamin,dzasum,dznrm2,
-    zgeadd, dzsum);
+    zgeadd, dzsum, zgemmt);
 
 @blasobjs = (lsame, xerbla);
 @bfblasobjs = (sbgemm, sbgemv, sbdot, sbstobf16, sbdtobf16, sbf16tos, dbf16tod);
@@ -60,7 +60,7 @@
     cblas_ctbsv, cblas_ctpmv, cblas_ctpsv, cblas_ctrmm, cblas_ctrmv, cblas_ctrsm, cblas_ctrsv, 
     cblas_scnrm2, cblas_scasum,
     cblas_icamax, cblas_icamin, cblas_icmin, cblas_icmax, cblas_scsum,cblas_cimatcopy,cblas_comatcopy
-    );
+    cblas_cgemmt);
 @cblasobjsd = (
     cblas_dasum, cblas_daxpy, cblas_dcopy, cblas_ddot,
     cblas_dgbmv, cblas_dgemm, cblas_dgemv, cblas_dger, cblas_dnrm2,
@@ -69,7 +69,7 @@
     cblas_dsyr2k, cblas_dsyr, cblas_dsyrk, cblas_dtbmv, cblas_dtbsv, cblas_dtpmv, cblas_dtpsv,
     cblas_dtrmm, cblas_dtrmv, cblas_dtrsm, cblas_dtrsv, cblas_daxpby, cblas_dgeadd,
     cblas_idamax, cblas_idamin, cblas_idmin, cblas_idmax, cblas_dsum,cblas_dimatcopy,cblas_domatcopy
-    );
+    cblas_dgemmt);
     
 @cblasobjss = (
     cblas_sasum, cblas_saxpy, cblas_saxpby,
@@ -80,7 +80,7 @@
     cblas_stbmv, cblas_stbsv, cblas_stpmv, cblas_stpsv, cblas_strmm, cblas_strmv, cblas_strsm,
     cblas_strsv, cblas_sgeadd,
     cblas_isamax, cblas_isamin, cblas_ismin, cblas_ismax, cblas_ssum,cblas_simatcopy,cblas_somatcopy
-    );
+    cblas_sgemmt);
 @cblasobjsz = (
     cblas_dzasum, cblas_dznrm2, cblas_zaxpy, cblas_zcopy, cblas_zdotc, cblas_zdotu, cblas_zdscal,
     cblas_zgbmv, cblas_zgemm, cblas_zgemv, cblas_zgerc, cblas_zgeru, cblas_zhbmv, cblas_zhemm,
@@ -90,7 +90,7 @@
     cblas_ztrsv, cblas_cdotc_sub, cblas_cdotu_sub, cblas_zdotc_sub, cblas_zdotu_sub,
     cblas_zaxpby, cblas_zgeadd,
     cblas_izamax, cblas_izamin, cblas_izmin, cblas_izmax, cblas_dzsum,cblas_zimatcopy,cblas_zomatcopy
-);
+    cblas_zgemmt);
 
 @cblasobjs = (  cblas_xerbla );
 

f_check

@@ -101,7 +101,14 @@ else
                         *flang*)
                             vendor=FLANG
                             openmp='-fopenmp'
-                            ;;
+			    data=`$compiler -v 2>&1 > /dev/null `
+                	    v="${data#*version *}"
+                	    v="${v%%*.}"
+                	    major="${v%%.*}"
+                	    if [ "$major" -ge 17 ]; then
+                        	vendor=FLANGNEW
+			    fi	
+			;;
                         *ifort*|*ifx*)
                             vendor=INTEL
                             openmp='-fopenmp'

getarch.c

@@ -1930,15 +1930,15 @@ printf("ELF_VERSION=2\n");
 
 #ifdef MAKE_NB_JOBS
   #if MAKE_NB_JOBS > 0
-    printf("MAKE += -j %d\n", MAKE_NB_JOBS);
+    printf("MAKEFLAGS += -j %d\n", MAKE_NB_JOBS);
   #else
     // Let make use parent -j argument or -j1 if there
     // is no make parent
   #endif
 #elif NO_PARALLEL_MAKE==1
-    printf("MAKE += -j 1\n");
+    printf("MAKEFLAGS += -j 1\n");
 #else
-    printf("MAKE += -j %d\n", get_num_cores());
+    printf("MAKEFLAGS += -j %d\n", get_num_cores());
 #endif
 
     break;

interface/geadd.c

@@ -68,7 +68,7 @@ void NAME(blasint *M, blasint *N, FLOAT *ALPHA, FLOAT *a, blasint *LDA,
   info = 0;
 
 
-  if (lda < MAX(1, m))	info = 6;
+  if (lda < MAX(1, m))	info = 5;
   if (ldc < MAX(1, m))	info = 8;
 
   if (n < 0)		info = 2;

interface/gemm.c

@@ -154,6 +154,23 @@ static size_t zgemm_small_kernel_b0[] = {
 #endif
 #endif
 
+#if defined(__linux__) && defined(__x86_64__) && defined(BFLOAT16)
+#define XFEATURE_XTILEDATA 18
+#define ARCH_REQ_XCOMP_PERM 0x1023
+static int openblas_amxtile_permission = 0;
+static int init_amxtile_permission() {
+  long status =
+      syscall(SYS_arch_prctl, ARCH_REQ_XCOMP_PERM, XFEATURE_XTILEDATA);
+  if (status != 0) {
+    fprintf(stderr, "XTILEDATA permission not granted in your device(Linux, "
+                    "Intel Sapphier Rapids), skip sbgemm calculation\n");
+    return -1;
+  }
+  openblas_amxtile_permission = 1;
+  return 0;
+}
+#endif
+
 #ifndef CBLAS
 
 void NAME(char *TRANSA, char *TRANSB,
@@ -455,6 +472,20 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_TRANSPOSE TransA, enum CBLAS_TRANS
 
 #endif
 
+#if defined(__linux__) && defined(__x86_64__) && defined(BFLOAT16)
+#if defined(DYNAMIC_ARCH)
+  if (gotoblas->need_amxtile_permission &&
+      openblas_amxtile_permission == 0 && init_amxtile_permission() == -1) {
+    return;
+  }
+#endif
+#if !defined(DYNAMIC_ARCH) && defined(SAPPHIRERAPIDS)
+  if (openblas_amxtile_permission == 0 && init_amxtile_permission() == -1) {
+    return;
+  }
+#endif
+#endif  // defined(__linux__) && defined(__x86_64__) && defined(BFLOAT16)
+
   if ((args.m == 0) || (args.n == 0)) return;
 
 #if 0

interface/gemmt.c

@@ -35,29 +35,26 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include "common.h"
-#ifdef FUNCTION_PROFILE
-#include "functable.h"
-#endif
 
 #ifndef COMPLEX
 #define SMP_THRESHOLD_MIN 65536.0
 #ifdef XDOUBLE
-#define ERROR_NAME "QGEMT "
+#define ERROR_NAME "QGEMMT "
 #elif defined(DOUBLE)
-#define ERROR_NAME "DGEMT "
+#define ERROR_NAME "DGEMMT "
 #elif defined(BFLOAT16)
-#define ERROR_NAME "SBGEMT "
+#define ERROR_NAME "SBGEMMT "
 #else
-#define ERROR_NAME "SGEMT "
+#define ERROR_NAME "SGEMMT "
 #endif
 #else
 #define SMP_THRESHOLD_MIN 8192.0
 #ifdef XDOUBLE
-#define ERROR_NAME "XGEMT "
+#define ERROR_NAME "XGEMMT "
 #elif defined(DOUBLE)
-#define ERROR_NAME "ZGEMT "
+#define ERROR_NAME "ZGEMMT "
 #else
-#define ERROR_NAME "CGEMT "
+#define ERROR_NAME "CGEMMT "
 #endif
 #endif
 
@@ -68,18 +65,19 @@
 #ifndef CBLAS
 
 void NAME(char *UPLO, char *TRANSA, char *TRANSB,
-	  blasint * M, blasint * N, blasint * K,
+	  blasint * M, blasint * K,
 	  FLOAT * Alpha,
 	  IFLOAT * a, blasint * ldA,
 	  IFLOAT * b, blasint * ldB, FLOAT * Beta, FLOAT * c, blasint * ldC)
 {
 
-	blasint m, n, k;
+	blasint m, k;
 	blasint lda, ldb, ldc;
 	int transa, transb, uplo;
 	blasint info;
 
 	char transA, transB, Uplo;
+	blasint nrowa, nrowb;
 	IFLOAT *buffer;
 	IFLOAT *aa, *bb;
 	FLOAT *cc;
@@ -92,7 +90,6 @@ void NAME(char *UPLO, char *TRANSA, char *TRANSB,
 	PRINT_DEBUG_NAME;
 
 	m = *M;
-	n = *N;
 	k = *K;
 
 #if defined(COMPLEX)
@@ -159,32 +156,39 @@ void NAME(char *UPLO, char *TRANSA, char *TRANSB,
 	if (Uplo == 'L')
 		uplo = 1;
 
+	nrowa = m;
+	if (transa) nrowa = k;
+	nrowb = k;
+	if (transb) nrowb = m;
+
 	info = 0;
 
-	if (uplo < 0)
-		info = 14;
-	if (ldc < m)
+	if (ldc < MAX(1, m))
 		info = 13;
+	if (ldb < MAX(1, nrowa))
+		info = 10;
+	if (lda < MAX(1, nrowb))
+		info = 8;
 	if (k < 0)
 		info = 5;
-	if (n < 0)
-		info = 4;
 	if (m < 0)
-		info = 3;
+		info = 4;
 	if (transb < 0)
-		info = 2;
+		info = 3;
 	if (transa < 0)
+		info = 2;
+	if (uplo < 0)
 		info = 1;
 
-	if (info) {
+	if (info != 0) {
 		BLASFUNC(xerbla) (ERROR_NAME, &info, sizeof(ERROR_NAME));
 		return;
 	}
 #else
 
 void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,
-	   enum CBLAS_TRANSPOSE TransA, enum CBLAS_TRANSPOSE TransB, blasint M,
-	   blasint N, blasint k,
+	   enum CBLAS_TRANSPOSE TransA, enum CBLAS_TRANSPOSE TransB, blasint m,
+	   blasint k,
 #ifndef COMPLEX
 	   FLOAT alpha,
 	   IFLOAT * A, blasint LDA,
@@ -205,17 +209,20 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,
 
 	int transa, transb, uplo;
 	blasint info;
-	blasint m, n, lda, ldb;
+	blasint lda, ldb;
 	FLOAT *a, *b;
 	XFLOAT *buffer;
 
 	PRINT_DEBUG_CNAME;
 
+	uplo = -1;
 	transa = -1;
 	transb = -1;
 	info = 0;
 
 	if (order == CblasColMajor) {
+		if (Uplo == CblasUpper) uplo = 0;
+		if (Uplo == CblasLower) uplo = 1;
 
 		if (TransA == CblasNoTrans)
 			transa = 0;
@@ -248,9 +255,6 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,
 			transb = 3;
 #endif
 
-		m = M;
-		n = N;
-
 		a = (void *)A;
 		b = (void *)B;
 		lda = LDA;
@@ -258,23 +262,31 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,
 
 		info = -1;
 
-		if (ldc < m)
+		blasint nrowa, nrowb;
+		nrowa = m;
+		if (transa) nrowa = k;
+		nrowb = k;
+		if (transb) nrowb = m;
+
+		if (ldc < MAX(1, m))
 			info = 13;
+		if (ldb < MAX(1, nrowb))
+			info = 10;
+		if (lda < MAX(1, nrowa))
+			info = 8;
 		if (k < 0)
 			info = 5;
-		if (n < 0)
-			info = 4;
 		if (m < 0)
-			info = 3;
+			info = 4;
 		if (transb < 0)
-			info = 2;
+			info = 3;
 		if (transa < 0)
+			info = 2;
+		if (uplo < 0)
 			info = 1;
 	}
 
 	if (order == CblasRowMajor) {
-		m = N;
-		n = M;
 
 		a = (void *)B;
 		b = (void *)A;
@@ -282,6 +294,9 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,
 		lda = LDB;
 		ldb = LDA;
 
+		if (Uplo == CblasUpper) uplo = 0;
+		if (Uplo == CblasLower) uplo = 1;
+
 		if (TransB == CblasNoTrans)
 			transa = 0;
 		if (TransB == CblasTrans)
@@ -315,29 +330,30 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,
 
 		info = -1;
 
-		if (ldc < m)
+		blasint ncola, ncolb;
+		ncola = k;
+		if (transa) ncola = m;
+		ncolb = m;
+		if (transb) ncolb = k;
+
+		if (ldc < MAX(1,m))
 			info = 13;
+		if (ldb < MAX(1, ncolb))
+			info = 10;
+		if (lda < MAX(1, ncola))
+			info = 8;
 		if (k < 0)
 			info = 5;
-		if (n < 0)
-			info = 4;
 		if (m < 0)
-			info = 3;
+			info = 4;
 		if (transb < 0)
-			info = 2;
+			info = 3;
 		if (transa < 0)
+			info = 2;
+		if (uplo < 0)
 			info = 1;
-
 	}
 
-	uplo = -1;
-	if (Uplo == CblasUpper)
-		uplo = 0;
-	if (Uplo == CblasLower)
-		uplo = 1;
-	if (uplo < 0)
-		info = 14;
-
 	if (info >= 0) {
 		BLASFUNC(xerbla) (ERROR_NAME, &info, sizeof(ERROR_NAME));
 		return;
@@ -407,37 +423,35 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,
 
 #endif
 
-	if ((m == 0) || (n == 0))
+	if (m == 0)
 		return;
 
 	IDEBUG_START;
 
-	FUNCTION_PROFILE_START();
-
 	const blasint incb = (transb == 0) ? 1 : ldb;
 
 	if (uplo == 1) {
-		for (i = 0; i < n; i++) {
-			j = n - i;
+		for (i = 0; i < m; i++) {
+			j = m - i;
 
 			l = j;
 #if defined(COMPLEX)
 			aa = a + i * 2;
 			bb = b + i * ldb * 2;
 			if (transa) {
-				l = k;
 				aa = a + lda * i * 2;
-				bb = b + i * 2;
 			}
+			if (transb)
+				bb = b + i * 2;
 			cc = c + i * 2 * ldc + i * 2;
 #else
 			aa = a + i;
 			bb = b + i * ldb;
 			if (transa) {
-				l = k;
 				aa = a + lda * i;
-				bb = b + i;
 			}
+			if (transb)
+				bb = b + i;
 			cc = c + i * ldc + i;
 #endif
 
@@ -458,8 +472,6 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,
 
 			IDEBUG_START;
 
-			FUNCTION_PROFILE_START();
-
 			buffer_size = j + k + 128 / sizeof(FLOAT);
 #ifdef WINDOWS_ABI
 			buffer_size += 160 / sizeof(FLOAT);
@@ -479,20 +491,34 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,
 #endif
 
 #if defined(COMPLEX)
+				if (!transa)
 				(gemv[(int)transa]) (j, k, 0, alpha_r, alpha_i,
 						     aa, lda, bb, incb, cc, 1,
 						     buffer);
+				else
+				(gemv[(int)transa]) (k, j, 0, alpha_r, alpha_i,
+						     aa, lda, bb, incb, cc, 1,
+						     buffer);
 #else
+				if (!transa)
 				(gemv[(int)transa]) (j, k, 0, alpha, aa, lda,
 						     bb, incb, cc, 1, buffer);
+				else
+				(gemv[(int)transa]) (k, j, 0, alpha, aa, lda,
+						     bb, incb, cc, 1, buffer);
 #endif
 #ifdef SMP
 			} else {
-
+				if (!transa)
 				(gemv_thread[(int)transa]) (j, k, alpha, aa,
 							    lda, bb, incb, cc,
 							    1, buffer,
 							    nthreads);
+				else
+				(gemv_thread[(int)transa]) (k, j, alpha, aa,
+							    lda, bb, incb, cc,
+							    1, buffer,
+							    nthreads);
 
 			}
 #endif
@@ -501,21 +527,19 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,
 		}
 	} else {
 
-		for (i = 0; i < n; i++) {
+		for (i = 0; i < m; i++) {
 			j = i + 1;
 
 			l = j;
 #if defined COMPLEX
 			bb = b + i * ldb * 2;
-			if (transa) {
-				l = k;
+			if (transb) {
 				bb = b + i * 2;
 			}
 			cc = c + i * 2 * ldc;
 #else
 			bb = b + i * ldb;
-			if (transa) {
-				l = k;
+			if (transb) {
 				bb = b + i;
 			}
 			cc = c + i * ldc;
@@ -537,8 +561,6 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,
 #endif
 			IDEBUG_START;
 
-			FUNCTION_PROFILE_START();
-
 			buffer_size = j + k + 128 / sizeof(FLOAT);
 #ifdef WINDOWS_ABI
 			buffer_size += 160 / sizeof(FLOAT);
@@ -558,30 +580,39 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,
 #endif
 
 #if defined(COMPLEX)
+				if (!transa)
 				(gemv[(int)transa]) (j, k, 0, alpha_r, alpha_i,
 						     a, lda, bb, incb, cc, 1,
 						     buffer);
+				else
+				(gemv[(int)transa]) (k, j, 0, alpha_r, alpha_i,
+						     a, lda, bb, incb, cc, 1,
+						     buffer);
 #else
+				if (!transa)
 				(gemv[(int)transa]) (j, k, 0, alpha, a, lda, bb,
 						     incb, cc, 1, buffer);
+				else
+				(gemv[(int)transa]) (k, j, 0, alpha, a, lda, bb,
+						     incb, cc, 1, buffer);
 #endif
 
 #ifdef SMP
 			} else {
-
+				if (!transa)
 				(gemv_thread[(int)transa]) (j, k, alpha, a, lda,
 							    bb, incb, cc, 1,
 							    buffer, nthreads);
-
+				else
+				(gemv_thread[(int)transa]) (k, j, alpha, a, lda,
+							    bb, incb, cc, 1,
+							    buffer, nthreads);
 			}
 #endif
 
 			STACK_FREE(buffer);
 		}
 	}
-	FUNCTION_PROFILE_END(COMPSIZE * COMPSIZE,
-			     args.m * args.k + args.k * args.n +
-			     args.m * args.n, 2 * args.m * args.n * args.k);
 
 	IDEBUG_END;
 

interface/imatcopy.c

@@ -100,13 +100,13 @@ void CNAME( enum CBLAS_ORDER CORDER, enum CBLAS_TRANSPOSE CTRANS, blasint crows,
 
 	if ( order == BlasColMajor)
 	{
-        	if ( trans == BlasNoTrans  &&  *ldb < *rows ) info = 9;
-        	if ( trans == BlasTrans    &&  *ldb < *cols ) info = 9;
+        	if ( trans == BlasNoTrans  &&  *ldb < *rows ) info = 8;
+        	if ( trans == BlasTrans    &&  *ldb < *cols ) info = 8;
 	}
 	if ( order == BlasRowMajor)
 	{
-        	if ( trans == BlasNoTrans  &&  *ldb < *cols ) info = 9;
-        	if ( trans == BlasTrans    &&  *ldb < *rows ) info = 9;
+        	if ( trans == BlasNoTrans  &&  *ldb < *cols ) info = 8;
+        	if ( trans == BlasTrans    &&  *ldb < *rows ) info = 8;
 	}
 
 	if ( order == BlasColMajor &&  *lda < *rows ) info = 7;
@@ -120,17 +120,20 @@ void CNAME( enum CBLAS_ORDER CORDER, enum CBLAS_TRANSPOSE CTRANS, blasint crows,
     		BLASFUNC(xerbla)(ERROR_NAME, &info, sizeof(ERROR_NAME));
     		return;
   	}
+
 #ifdef NEW_IMATCOPY
-    if ( *lda == *ldb && *rows == *cols) {
+    if ( *lda == *ldb ) {
         if ( order == BlasColMajor )
         {
             if ( trans == BlasNoTrans )
             {
                 IMATCOPY_K_CN(*rows, *cols, *alpha, a, *lda );
+                return;
             }
-            else
+            else if ( *rows == *cols )
             {
                 IMATCOPY_K_CT(*rows, *cols, *alpha, a, *lda );
+                return;
             }
         }
         else
@@ -138,26 +141,23 @@ void CNAME( enum CBLAS_ORDER CORDER, enum CBLAS_TRANSPOSE CTRANS, blasint crows,
             if ( trans == BlasNoTrans )
             {
                 IMATCOPY_K_RN(*rows, *cols, *alpha, a, *lda );
+                return;
             }
-            else
+            else if ( *rows == *cols )
             {
                 IMATCOPY_K_RT(*rows, *cols, *alpha, a, *lda );
+                return;
             }
         }
-        return; 
     }
-
 #endif
 
-	if ( *lda >  *ldb )
-		msize = (size_t)(*lda) * (*ldb)  * sizeof(FLOAT);
-	else
-		msize = (size_t)(*ldb) * (*ldb)  * sizeof(FLOAT);
+	msize = (size_t)(*rows) * (*cols) * sizeof(FLOAT);
 
 	b = malloc(msize);
 	if ( b == NULL )
 	{
-		printf("Memory alloc failed\n");
+		printf("Memory alloc failed in imatcopy\n");
 		exit(1);
 	}
 
@@ -165,26 +165,26 @@ void CNAME( enum CBLAS_ORDER CORDER, enum CBLAS_TRANSPOSE CTRANS, blasint crows,
 	{
 		if ( trans == BlasNoTrans )
 		{
-			OMATCOPY_K_CN(*rows, *cols, *alpha, a, *lda, b, *ldb );
-			OMATCOPY_K_CN(*rows, *cols, (FLOAT) 1.0 , b, *ldb, a, *ldb );
+			OMATCOPY_K_CN(*rows, *cols, *alpha, a, *lda, b, *rows );
+			OMATCOPY_K_CN(*rows, *cols, (FLOAT) 1.0 , b, *rows, a, *ldb );
 		}
 		else
 		{
-			OMATCOPY_K_CT(*rows, *cols, *alpha, a, *lda, b, *ldb );
-			OMATCOPY_K_CN(*cols, *rows, (FLOAT) 1.0, b, *ldb, a, *ldb );
+			OMATCOPY_K_CT(*rows, *cols, *alpha, a, *lda, b, *cols );
+			OMATCOPY_K_CN(*cols, *rows, (FLOAT) 1.0, b, *cols, a, *ldb );
 		}
 	}
 	else
 	{
 		if ( trans == BlasNoTrans )
 		{
-			OMATCOPY_K_RN(*rows, *cols, *alpha, a, *lda, b, *ldb );
-			OMATCOPY_K_RN(*rows, *cols, (FLOAT) 1.0, b, *ldb, a, *ldb );
+			OMATCOPY_K_RN(*rows, *cols, *alpha, a, *lda, b, *cols );
+			OMATCOPY_K_RN(*rows, *cols, (FLOAT) 1.0, b, *cols, a, *ldb );
 		}
 		else
 		{
-			OMATCOPY_K_RT(*rows, *cols, *alpha, a, *lda, b, *ldb );
-			OMATCOPY_K_RN(*cols, *rows, (FLOAT) 1.0, b, *ldb, a, *ldb );
+			OMATCOPY_K_RT(*rows, *cols, *alpha, a, *lda, b, *rows );
+			OMATCOPY_K_RN(*cols, *rows, (FLOAT) 1.0, b, *rows, a, *ldb );
 		}
 	}
 

interface/nrm2.c

@@ -54,6 +54,21 @@ FLOATRET NAME(blasint *N, FLOAT *x, blasint *INCX){
 
   if (n <= 0) return 0.;
 
+#ifndef COMPLEX
+  if (n == 1)
+#ifdef DOUBLE
+    return fabs(x[0]);
+#else
+    return fabsf(x[0]);
+#endif
+#endif
+
+  if (incx < 0) 
+#ifdef COMPLEX    
+    x -= (n - 1) * incx * 2;
+#else
+    x -= (n - 1) * incx;
+#endif
   IDEBUG_START;
 
   FUNCTION_PROFILE_START();
@@ -82,6 +97,22 @@ FLOAT CNAME(blasint n, FLOAT *x, blasint incx){
 
   if (n <= 0) return 0.;
 
+  #ifndef COMPLEX
+  if (n == 1)
+#ifdef DOUBLE
+    return fabs(x[0]);
+#else
+    return fabsf(x[0]);
+#endif
+#endif
+
+  if (incx < 0) 
+#ifdef COMPLEX    
+    x -= (n - 1) * incx * 2;
+#else
+    x -= (n - 1) * incx;
+#endif
+  
   IDEBUG_START;
 
   FUNCTION_PROFILE_START();

interface/rotg.c

@@ -1,9 +1,11 @@
 #include <math.h>
+#include <float.h>
 #include "common.h"
 #ifdef FUNCTION_PROFILE
 #include "functable.h"
 #endif
 
+
 #ifndef CBLAS
 
 void NAME(FLOAT *DA, FLOAT *DB, FLOAT *C, FLOAT *S){
@@ -14,17 +16,27 @@ void CNAME(FLOAT *DA, FLOAT *DB, FLOAT *C, FLOAT *S){
 
 #endif
 
+#ifdef DOUBLE
+  long double safmin = DBL_MIN;
+#else
+  long double safmin = FLT_MIN;
+#endif
+
 #if defined(__i386__) || defined(__x86_64__) || defined(__ia64__) || defined(_M_X64) || defined(_M_IX86)
 
   long double da = *DA;
   long double db = *DB;
   long double c;
   long double s;
-  long double r, roe, z;
+  long double r, z;
+  long double sigma, dascal,dbscal;
 
   long double ada = fabsl(da);
   long double adb = fabsl(db);
-  long double scale = ada + adb;
+  long double maxab = MAX(ada,adb);
+  long double safmax;
+  long double scale;
+
 
 #ifndef CBLAS
   PRINT_DEBUG_NAME;
@@ -32,17 +44,25 @@ void CNAME(FLOAT *DA, FLOAT *DB, FLOAT *C, FLOAT *S){
   PRINT_DEBUG_CNAME;
 #endif
 
-  roe = db;
-  if (ada > adb) roe = da;
-
-  if (scale == ZERO) {
+  if (adb == ZERO) {
     *C = ONE;
     *S = ZERO;
-    *DA = ZERO;
     *DB = ZERO;
+  } else if (ada == ZERO) {
+    *C = ZERO;
+    *S = ONE;
+    *DA = *DB;
+    *DB = ONE;
   } else {
-    r = sqrt(da * da + db * db);
-    if (roe < 0) r = -r;
+  safmax = 1./safmin;
+  scale = MIN(MAX(safmin,maxab), safmax);
+    if (ada > adb)
+	sigma = copysign(1.,da);
+    else
+	sigma = copysign(1.,db);
+    dascal = da / scale;
+    dbscal = db / scale;
+    r = sigma * (scale * sqrt(dascal * dascal + dbscal * dbscal));
     c = da / r;
     s = db / r;
     z = ONE;
@@ -65,11 +85,22 @@ void CNAME(FLOAT *DA, FLOAT *DB, FLOAT *C, FLOAT *S){
   FLOAT db = *DB;
   FLOAT c  = *C;
   FLOAT s  = *S;
-  FLOAT r, roe, z;
+  FLOAT sigma;
+  FLOAT r, z;
 
   FLOAT ada = fabs(da);
   FLOAT adb = fabs(db);
-  FLOAT scale = ada + adb;
+  FLOAT maxab = MAX(ada,adb);
+  long double safmax ;
+  FLOAT scale ;
+
+  safmax = 1./safmin;
+  scale = MIN(MAX(safmin,maxab), safmax);
+
+  if (ada > adb)
+	sigma = copysign(1.,da);
+    else
+	sigma = copysign(1.,db);
 
 #ifndef CBLAS
   PRINT_DEBUG_NAME;
@@ -77,20 +108,21 @@ void CNAME(FLOAT *DA, FLOAT *DB, FLOAT *C, FLOAT *S){
   PRINT_DEBUG_CNAME;
 #endif
 
-  roe = db;
-  if (ada > adb) roe = da;
 
-  if (scale == ZERO) {
+  if (adb == ZERO) {
     *C = ONE;
     *S = ZERO;
-    *DA = ZERO;
     *DB = ZERO;
+  } else if (ada == ZERO) {
+    *C = ZERO;
+    *S = ONE;
+    *DA = *DB;
+    *DB = ONE;
   } else {
     FLOAT aa = da / scale;
     FLOAT bb = db / scale;
 
-    r = scale * sqrt(aa * aa + bb * bb);
-    if (roe < 0) r = -r;
+    r = sigma * scale * sqrt(aa * aa + bb * bb);
     c = da / r;
     s = db / r;
     z = ONE;

interface/symm.c

@@ -166,7 +166,7 @@ void NAME(char *SIDE, char *UPLO,
   int nodes;
 #endif
 # if defined(SMP)
-  int MN;
+  double MN;
 #endif
   blasint info;
   int side;
@@ -264,7 +264,7 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_SIDE Side, enum CBLAS_UPLO Uplo,
   int nodes;
 #endif
 #if defined(SMP)
-  int MN;
+  double MN;
 #endif
 
   PRINT_DEBUG_CNAME;

interface/syrk.c

@@ -107,7 +107,7 @@ void NAME(char *UPLO, char *TRANS,
   FLOAT *sa, *sb;
 
 #ifdef SMP
-  int NNK;
+  double NNK;
 #ifdef USE_SIMPLE_THREADED_LEVEL3
 #ifndef COMPLEX
 #ifdef XDOUBLE
@@ -232,7 +232,7 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo, enum CBLAS_TRANSPOSE Tr
   FLOAT *sa, *sb;
 
 #ifdef SMP
-int NNK;
+double NNK;
 
 #ifdef USE_SIMPLE_THREADED_LEVEL3
 #ifndef COMPLEX

interface/zimatcopy.c

@@ -125,27 +125,33 @@ void CNAME( enum CBLAS_ORDER CORDER, enum CBLAS_TRANSPOSE CTRANS, blasint crows,
     		BLASFUNC(xerbla)(ERROR_NAME, &info, sizeof(ERROR_NAME));
     		return;
   	}
+
 #ifdef NEW_IMATCOPY
-    if (*lda == *ldb && *cols == *rows) {
+    if (*lda == *ldb ) {
         if ( order == BlasColMajor )
         {
 
             if ( trans == BlasNoTrans )
             {
                 IMATCOPY_K_CN(*rows, *cols, alpha[0], alpha[1], a, *lda );
+                return;
             }
             if ( trans == BlasConj )
             {
                 IMATCOPY_K_CNC(*rows, *cols, alpha[0], alpha[1], a, *lda );
+                return;
             }
-            if ( trans == BlasTrans )
+            if ( trans == BlasTrans && *rows == *cols )
             {
                 IMATCOPY_K_CT(*rows, *cols, alpha[0], alpha[1], a, *lda );
+                return;
             }
-            if ( trans == BlasTransConj )
+            if ( trans == BlasTransConj && *rows == *cols )
             {
                 IMATCOPY_K_CTC(*rows, *cols, alpha[0], alpha[1], a, *lda );
+                return;
             }
+
         }
         else
         {
@@ -153,67 +159,59 @@ void CNAME( enum CBLAS_ORDER CORDER, enum CBLAS_TRANSPOSE CTRANS, blasint crows,
             if ( trans == BlasNoTrans )
             {
                 IMATCOPY_K_RN(*rows, *cols, alpha[0], alpha[1], a, *lda );
+                return;
             }
             if ( trans == BlasConj )
             {
                 IMATCOPY_K_RNC(*rows, *cols, alpha[0], alpha[1], a, *lda );
+                return;
             }
-            if ( trans == BlasTrans )
+            if ( trans == BlasTrans && *rows == *cols )
             {
                 IMATCOPY_K_RT(*rows, *cols, alpha[0], alpha[1], a, *lda );
+                return;
             }
-            if ( trans == BlasTransConj )
+            if ( trans == BlasTransConj && *rows == *cols )
             {
                 IMATCOPY_K_RTC(*rows, *cols, alpha[0], alpha[1], a, *lda );
+                return;
             }
+
         }
-        return; 
     }
 #endif
 
-	if ( *lda >  *ldb )
-                msize = (size_t)(*lda) * (*ldb)  * sizeof(FLOAT) * 2;
-        else
-                msize = (size_t)(*ldb) * (*ldb)  * sizeof(FLOAT) * 2;
-
-        b = malloc(msize);
-        if ( b == NULL )
-        {
-                printf("Memory alloc failed in zimatcopy\n");
-                exit(1);
-        }
+	msize = (size_t)(*rows) * (*cols) * sizeof(FLOAT) * 2;
 
+	b = malloc(msize);
+	if ( b == NULL )
+	{
+		printf("Memory alloc failed in zimatcopy\n");
+		exit(1);
+	}
 
 	if ( order == BlasColMajor )
 	{
 
 		if ( trans == BlasNoTrans )
 		{
-	  		OMATCOPY_K_CN(*rows, *cols, alpha[0], alpha[1], a, *lda, b, *ldb );
-	  		OMATCOPY_K_CN(*rows, *cols, (FLOAT) 1.0, (FLOAT) 0.0 , b, *ldb, a, *ldb );
-			free(b);
-			return;
+	  		OMATCOPY_K_CN(*rows, *cols, alpha[0], alpha[1], a, *lda, b, *rows );
+	  		OMATCOPY_K_CN(*rows, *cols, (FLOAT) 1.0, (FLOAT) 0.0 , b, *rows, a, *ldb );
 		}
-		if ( trans == BlasConj )
+		else if ( trans == BlasConj )
 		{
-			OMATCOPY_K_CNC(*rows, *cols, alpha[0], alpha[1], a, *lda, b, *ldb );
-	  		OMATCOPY_K_CN(*rows, *cols, (FLOAT) 1.0, (FLOAT) 0.0 , b, *ldb, a, *ldb );
-			free(b);
-			return;
+			OMATCOPY_K_CNC(*rows, *cols, alpha[0], alpha[1], a, *lda, b, *rows );
+	  		OMATCOPY_K_CN(*rows, *cols, (FLOAT) 1.0, (FLOAT) 0.0 , b, *rows, a, *ldb );
 		}
-		if ( trans == BlasTrans )
+		else if ( trans == BlasTrans )
 		{
-			OMATCOPY_K_CT(*rows, *cols, alpha[0], alpha[1], a, *lda, b, *ldb );
-	  		OMATCOPY_K_CN(*cols, *rows, (FLOAT) 1.0, (FLOAT) 0.0 , b, *ldb, a, *ldb );
-			free(b);
-			return;
+			OMATCOPY_K_CT(*rows, *cols, alpha[0], alpha[1], a, *lda, b, *cols );
+	  		OMATCOPY_K_CN(*cols, *rows, (FLOAT) 1.0, (FLOAT) 0.0 , b, *cols, a, *ldb );
 		}
-		if ( trans == BlasTransConj )
+		else if ( trans == BlasTransConj )
 		{
-			OMATCOPY_K_CTC(*rows, *cols, alpha[0], alpha[1], a, *lda, b, *ldb );
-	  		OMATCOPY_K_CN(*cols, *rows, (FLOAT) 1.0, (FLOAT) 0.0 , b, *ldb, a, *ldb );
-			free(b);
-			return;
+			OMATCOPY_K_CTC(*rows, *cols, alpha[0], alpha[1], a, *lda, b, *cols );
+	  		OMATCOPY_K_CN(*cols, *rows, (FLOAT) 1.0, (FLOAT) 0.0 , b, *cols, a, *ldb );
 		}
 
 	}
@@ -222,34 +220,27 @@ void CNAME( enum CBLAS_ORDER CORDER, enum CBLAS_TRANSPOSE CTRANS, blasint crows,
 
 		if ( trans == BlasNoTrans )
 		{
-			OMATCOPY_K_RN(*rows, *cols, alpha[0], alpha[1], a, *lda, b, *ldb );
-	  		OMATCOPY_K_RN(*rows, *cols, (FLOAT) 1.0, (FLOAT) 0.0 , b, *ldb, a, *ldb );
-			free(b);
-			return;
+			OMATCOPY_K_RN(*rows, *cols, alpha[0], alpha[1], a, *lda, b, *cols );
+	  		OMATCOPY_K_RN(*rows, *cols, (FLOAT) 1.0, (FLOAT) 0.0 , b, *cols, a, *ldb );
 		}
-		if ( trans == BlasConj )
+		else if ( trans == BlasConj )
 		{
-			OMATCOPY_K_RNC(*rows, *cols, alpha[0], alpha[1], a, *lda, b, *ldb );
-	  		OMATCOPY_K_RN(*rows, *cols, (FLOAT) 1.0, (FLOAT) 0.0 , b, *ldb, a, *ldb );
-			free(b);
-			return;
+			OMATCOPY_K_RNC(*rows, *cols, alpha[0], alpha[1], a, *lda, b, *cols );
+	  		OMATCOPY_K_RN(*rows, *cols, (FLOAT) 1.0, (FLOAT) 0.0 , b, *cols, a, *ldb );
 		}
-		if ( trans == BlasTrans )
+		else if ( trans == BlasTrans )
 		{
-			OMATCOPY_K_RT(*rows, *cols, alpha[0], alpha[1], a, *lda, b, *ldb );
-	  		OMATCOPY_K_RN(*cols, *rows, (FLOAT) 1.0, (FLOAT) 0.0 , b, *ldb, a, *ldb );
-			free(b);
-			return;
+			OMATCOPY_K_RT(*rows, *cols, alpha[0], alpha[1], a, *lda, b, *rows );
+	  		OMATCOPY_K_RN(*cols, *rows, (FLOAT) 1.0, (FLOAT) 0.0 , b, *rows, a, *ldb );
 		}
-		if ( trans == BlasTransConj )
+		else if ( trans == BlasTransConj )
 		{
-			OMATCOPY_K_RTC(*rows, *cols, alpha[0], alpha[1], a, *lda, b, *ldb );
-	  		OMATCOPY_K_RN(*cols, *rows, (FLOAT) 1.0, (FLOAT) 0.0 , b, *ldb, a, *ldb );
-			free(b);
-			return;
+			OMATCOPY_K_RTC(*rows, *cols, alpha[0], alpha[1], a, *lda, b, *rows );
+	  		OMATCOPY_K_RN(*cols, *rows, (FLOAT) 1.0, (FLOAT) 0.0 , b, *rows, a, *ldb );
 		}
 
 	}
+
 	free(b);
 	return;
 

interface/zrotg.c

@@ -1,9 +1,11 @@
 #include <math.h>
+#include <float.h>
 #include "common.h"
 #ifdef FUNCTION_PROFILE
 #include "functable.h"
 #endif
 
+
 #ifndef CBLAS
 void NAME(FLOAT *DA, FLOAT *DB, FLOAT *C, FLOAT *S){
 
@@ -14,53 +16,28 @@ void CNAME(void *VDA, void *VDB, FLOAT *C, void *VS) {
     FLOAT *S  = (FLOAT*) VS;
 #endif /* CBLAS */
 
-#if defined(__i386__) || defined(__x86_64__) || defined(__ia64__) || defined(_M_X64) || defined(_M_IX86)
-
-  long double da_r = *(DA + 0);
-  long double da_i = *(DA + 1);
-  long double db_r = *(DB + 0);
-  long double db_i = *(DB + 1);
-  long double r;
-
-  long double ada = fabsl(da_r) + fabsl(da_i);
-
-  PRINT_DEBUG_NAME;
-
-  IDEBUG_START;
-
-  FUNCTION_PROFILE_START();
-
-  if (ada == ZERO) {
-    *C        = ZERO;
-    *(S  + 0) = ONE;
-    *(S  + 1) = ZERO;
-    *(DA + 0) = db_r;
-    *(DA + 1) = db_i;
-  } else {
-    long double alpha_r, alpha_i;
-
-    ada = sqrt(da_r * da_r + da_i * da_i);
-
-    r = sqrt(da_r * da_r + da_i * da_i + db_r * db_r + db_i * db_i);
+#ifdef DOUBLE
+  long double safmin = DBL_MIN;
+  long double rtmin = sqrt(DBL_MIN/DBL_EPSILON);
+#else
+  long double safmin = FLT_MIN;
+  long double rtmin = sqrt(FLT_MIN/FLT_EPSILON);
+#endif
 
-    alpha_r = da_r / ada;
-    alpha_i = da_i / ada;
 
-    *(C + 0)  = ada / r;
-    *(S + 0)  = (alpha_r * db_r + alpha_i *db_i) / r;
-    *(S + 1)  = (alpha_i * db_r - alpha_r *db_i) / r;
-    *(DA + 0) = alpha_r * r;
-    *(DA + 1) = alpha_i * r;
-  }
-#else
-  FLOAT da_r = *(DA + 0);
-  FLOAT da_i = *(DA + 1);
-  FLOAT db_r = *(DB + 0);
-  FLOAT db_i = *(DB + 1);
-  FLOAT r;
+  FLOAT da_r = *(DA+0);
+  FLOAT da_i = *(DA+1);
+  FLOAT db_r = *(DB+0);
+  FLOAT db_i = *(DB+1);
+  //long double r;
+  FLOAT *r, *S1=(FLOAT *)malloc(2*sizeof(FLOAT));
+  FLOAT *R=(FLOAT *)malloc(2*sizeof(FLOAT));
+  long double d;
 
-  FLOAT ada = fabs(da_r) + fabs(da_i);
-  FLOAT adb;
+  FLOAT ada =  da_r * da_r + da_i * da_i; 
+  FLOAT adb =  db_r * db_r + db_i * db_i; 
+  FLOAT adart = sqrt( da_r * da_r + da_i * da_i); 
+  FLOAT adbrt = sqrt( db_r * db_r + db_i * db_i); 
 
   PRINT_DEBUG_NAME;
 
@@ -68,69 +45,137 @@ void CNAME(void *VDA, void *VDB, FLOAT *C, void *VS) {
 
   FUNCTION_PROFILE_START();
 
-  if (ada == ZERO) {
-    *C        = ZERO;
-    *(S  + 0) = ONE;
+  if (db_r == ZERO && db_i == ZERO) {
+    *C        = ONE;
+    *(S  + 0) = ZERO;
     *(S  + 1) = ZERO;
-    *(DA + 0) = db_r;
-    *(DA + 1) = db_i;
-  } else {
-    FLOAT scale;
-    FLOAT aa_r, aa_i, bb_r, bb_i;
-    FLOAT alpha_r, alpha_i;
-
-    aa_r = fabs(da_r);
-    aa_i = fabs(da_i);
-
-    if (aa_i > aa_r) {
-      aa_r = fabs(da_i);
-      aa_i = fabs(da_r);
-    }
-
-    if (aa_r == ZERO) {
-	ada = 0.;
-    } else {
-        scale = (aa_i / aa_r);
-        ada = aa_r * sqrt(ONE + scale * scale);
-    }
-
-    bb_r = fabs(db_r);
-    bb_i = fabs(db_i);
-
-    if (bb_i > bb_r) {
-      bb_r = fabs(bb_i);
-      bb_i = fabs(bb_r);
-    }
-
-    if (bb_r == ZERO) {
-	adb = 0.;
-    } else {
-    	scale = (bb_i / bb_r);
-    	adb = bb_r * sqrt(ONE + scale * scale);
-    }
-    scale = ada + adb;
-
-    aa_r    = da_r / scale;
-    aa_i    = da_i / scale;
-    bb_r    = db_r / scale;
-    bb_i    = db_i / scale;
-
-    r = scale * sqrt(aa_r * aa_r + aa_i * aa_i + bb_r * bb_r + bb_i * bb_i);
-
-    alpha_r = da_r / ada;
-    alpha_i = da_i / ada;
-
-    *(C + 0)  = ada / r;
-    *(S + 0)  = (alpha_r * db_r + alpha_i *db_i) / r;
-    *(S + 1)  = (alpha_i * db_r - alpha_r *db_i) / r;
-    *(DA + 0) = alpha_r * r;
-    *(DA + 1) = alpha_i * r;
+    return;
   }
-#endif
 
-  FUNCTION_PROFILE_END(4, 4, 4);
-
-  IDEBUG_END;
-
-  return;
+  long double safmax = 1./safmin;
+#if defined DOUBLE
+  long double rtmax = safmax /DBL_EPSILON;
+#else
+  long double rtmax = safmax /FLT_EPSILON;
+#endif
+		*(S1 + 0) = *(DB + 0);
+		*(S1 + 1) = *(DB + 1) *-1;
+	if (da_r == ZERO && da_i == ZERO) {
+	    *C = ZERO;	  
+	    if (db_r == ZERO) {
+		    (*DA) = fabsl(db_i);
+		*S = *S1 /da_r;
+		*(S+1) = *(S1+1) /da_r;
+		return;
+	    } else if ( db_i == ZERO) {
+		    *DA = fabsl(db_r);
+		*S = *S1 /da_r;
+		*(S+1) = *(S1+1) /da_r;
+		return;
+	    } else {
+	        long double g1 = MAX( fabsl(db_r), fabsl(db_i));
+	        rtmax =sqrt(safmax/2.);
+	        if (g1 > rtmin && g1 < rtmax) { // unscaled
+		    d = sqrt(adb);
+		    *S = *S1 /d;
+		    *(S+1) = *(S1+1) /d;
+		    *DA = d ;
+		    *(DA+1) = ZERO;
+		    return;
+    	        } else { // scaled algorithm
+		  long double u = MIN ( safmax, MAX ( safmin, g1));
+		    FLOAT gs_r = db_r/u;
+		    FLOAT gs_i = db_i/u;
+		    d = sqrt ( gs_r*gs_r + gs_i*gs_i);
+		    *S = gs_r / d;
+		    *(S + 1) = (gs_i * -1) / d;
+		    *DA = d * u;
+		    *(DA+1) = ZERO;
+		    return;
+	        }
+	    }
+	} else {
+	       FLOAT f1 = MAX ( fabsl(da_r), fabsl(da_i));
+	       FLOAT g1 = MAX ( fabsl(db_r), fabsl(db_i));
+	       rtmax = sqrt(safmax / 4.);
+	       if ( f1 > rtmin && f1 < rtmax && g1 > rtmin && g1 < rtmax) { //unscaled
+		    long double h = ada + adb;
+	   	    double adahsq = sqrt(ada * h);
+		    if (ada >= h *safmin) {
+			*C = sqrt(ada/h);
+			*R = *DA / *C;
+			*(R+1) = *(DA+1) / *(C+1);
+			rtmax *= 2.;
+			if ( ada > rtmin && h < rtmax) { // no risk of intermediate overflow
+				*S = *S1 * (*DA / adahsq) - *(S1+1)* (*(DA+1)/adahsq);
+				*(S+1) = *S1 * (*(DA+1) / adahsq) + *(S1+1) * (*DA/adahsq);
+			} else {
+				*S = *S1 * (*R/h) - *(S1+1) * (*(R+1)/h);
+				*(S+1) = *S1 * (*(R+1)/h) + *(S1+1) * (*(R)/h);
+			}
+	    	    } else {
+		        *C = ada / adahsq;
+		        if (*C >= safmin) 
+			    *R = *DA / *C;
+		        else
+			    *R = *DA * (h / adahsq);
+		        *S = *S1 * ada / adahsq;
+		    	*(S+1) = *(S1+1) * ada / adahsq;
+		    }
+		    *DA=*R;
+		    *(DA+1)=*(R+1);
+		    return;
+	        } else {	// scaled
+		   FLOAT fs_r, fs_i, gs_r, gs_i;
+		long double v,w,f2,g2,h;
+		long double u = MIN ( safmax, MAX ( safmin, MAX(f1,g1)));
+		    gs_r = db_r/u;
+		    gs_i = db_i/u;
+		    g2 = sqrt ( gs_r*gs_r + gs_i*gs_i);
+		    if (f1 /u < rtmin) {
+			v = MIN (safmax, MAX (safmin, f1));
+			w = v / u;
+			fs_r = *DA/ v;
+			fs_i = *(DA+1) / v;
+		        f2 = sqrt ( fs_r*fs_r + fs_i*fs_i);
+		        h = f2 * w * w + g2;		
+		    } else { // use same scaling for both
+			w = 1.;
+			fs_r = *DA/ u;
+			fs_i = *(DA+1) / u;
+		        f2 = sqrt ( fs_r*fs_r + fs_i*fs_i);
+			h = f2 + g2;
+		    }
+		    if ( f2 >= h * safmin) {
+			    *C = sqrt ( f2 / h );
+			    *DA = fs_r / *C;
+			    *(DA+1) = fs_i / *C;
+			    rtmax *= 2;
+			    if ( f2 > rtmin && h < rtmax) {
+				    *S = gs_r * (fs_r /sqrt(f2*h)) - gs_i * (fs_i / sqrt(f2*h));
+				    *(S+1) = gs_r * (fs_i /sqrt(f2*h)) + gs_i * -1. * (fs_r / sqrt(f2*h));
+			    } else {
+				    *S = gs_r * (*DA/h) - gs_i * (*(DA+1) / h);
+				    *(S+1) = gs_r * (*(DA+1) /h) + gs_i * -1. * (*DA / h);
+			    }
+		    } else { // intermediates might overflow
+			d = sqrt ( f2 * h);
+			*C = f2 /d;
+			if (*C >= safmin) {
+				*DA = fs_r / *C;
+				*(DA+1) = fs_i / *C;
+			} else {
+				*DA = fs_r * (h / d);
+				*(DA+1) = fs_i / (h / d);
+			}
+			*S = gs_r * (fs_r /d) - gs_i * (fs_i / d);
+			*(S+1) = gs_r * (fs_i /d) + gs_i * -1. * (fs_r / d);
+	            }
+		    *C *= w;
+		    *DA *= u;
+		    *(DA+1) *= u;
+		    return;
+		}
+	}
 }
+	

kernel/Makefile

@@ -33,7 +33,7 @@ endif
 ifdef TARGET_CORE
 ifeq ($(TARGET_CORE), SAPPHIRERAPIDS)
  override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE)
- ifeq ($(GCCVERSIONGTEQ10), 1) 
+ ifeq (1, $(filter 1,$(GCCVERSIONGTEQ11) $(CLANGVERSIONGTEQ12)))
   override CFLAGS += -march=sapphirerapids
  else 
   override CFLAGS += -march=skylake-avx512 -mavx512f
@@ -48,7 +48,7 @@ ifeq ($(TARGET_CORE), SAPPHIRERAPIDS)
  endif
 else ifeq ($(TARGET_CORE), COOPERLAKE)
  override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE)
- ifeq ($(GCCVERSIONGTEQ10), 1) 
+ ifeq (1, $(filter 1,$(GCCVERSIONGTEQ10) $(CLANGVERSIONGTEQ9)))
   override CFLAGS += -march=cooperlake
  else 
   override CFLAGS += -march=skylake-avx512 -mavx512f
@@ -77,6 +77,12 @@ else ifeq ($(TARGET_CORE), ZEN)
  override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE) $(AVX2OPT)
 else ifeq ($(TARGET_CORE), LOONGSON3R4)
  override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE) $(MSA_FLAGS)
+else ifneq ($(filter NEOVERSEN2 NEOVERSEV1, $(TARGET_CORE)),)
+ ifeq ($(C_COMPILER), PGI)
+  override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE) -Msve_intrinsics
+ else
+  override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE)
+ endif
 else
  override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE)
 endif

kernel/Makefile.L3

@@ -35,6 +35,12 @@ USE_TRMM = 1
 endif
 endif
 
+ifneq ($(DYNAMIC_ARCH), 1)
+ifeq ($(TARGET), MIPS64_GENERIC)
+USE_TRMM = 1
+endif
+endif
+
 ifeq ($(CORE), HASWELL)
 USE_TRMM = 1
 endif

kernel/arm/nrm2.c

@@ -57,7 +57,7 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
 	FLOAT absxi = 0.0;
 
 
-	if (n <= 0 || inc_x <= 0) return(0.0);
+	if (n <= 0 || inc_x == 0) return(0.0);
 	if ( n == 1 ) return( ABS(x[0]) );
 
 	n *= inc_x;

kernel/arm/znrm2.c

@@ -57,7 +57,7 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
 	BLASLONG inc_x2;
 	FLOAT temp;
 
-	if (n <= 0 || inc_x <= 0) return(0.0);
+	if (n <= 0 || inc_x == 0) return(0.0);
 
 	inc_x2 = 2 * inc_x;
 

kernel/arm64/KERNEL.ARMV8SVE

@@ -57,7 +57,7 @@ CAMAXKERNEL  = zamax.S
 ZAMAXKERNEL  = zamax.S
 
 SAXPYKERNEL  = axpy.S
-DAXPYKERNEL  = axpy.S
+DAXPYKERNEL  = daxpy_thunderx2t99.S
 CAXPYKERNEL  = zaxpy.S
 ZAXPYKERNEL  = zaxpy.S
 
@@ -81,45 +81,35 @@ DGEMVTKERNEL = gemv_t.S
 CGEMVTKERNEL = zgemv_t.S
 ZGEMVTKERNEL = zgemv_t.S
 
-
-SASUMKERNEL    = asum.S
-DASUMKERNEL    = asum.S
-CASUMKERNEL    = casum.S
-ZASUMKERNEL    = zasum.S
-
-SCOPYKERNEL    = copy.S
-DCOPYKERNEL    = copy.S
-CCOPYKERNEL    = copy.S
-ZCOPYKERNEL    = copy.S
-
-SSWAPKERNEL    = swap.S
-DSWAPKERNEL    = swap.S
-CSWAPKERNEL    = swap.S
-ZSWAPKERNEL    = swap.S
-
-ISAMAXKERNEL   = iamax.S
-IDAMAXKERNEL   = iamax.S
-ICAMAXKERNEL   = izamax.S
-IZAMAXKERNEL   = izamax.S
-
-SNRM2KERNEL    = nrm2.S
-DNRM2KERNEL    = nrm2.S
-CNRM2KERNEL    = znrm2.S
-ZNRM2KERNEL    = znrm2.S
-
-DDOTKERNEL     = dot.S
-ifneq ($(C_COMPILER), PGI)
-SDOTKERNEL     = ../generic/dot.c
-else
-SDOTKERNEL = dot.S
-endif
-ifneq ($(C_COMPILER), PGI)
-CDOTKERNEL     = zdot.S
-ZDOTKERNEL     = zdot.S
-else
-CDOTKERNEL = ../arm/zdot.c
-ZDOTKERNEL = ../arm/zdot.c
-endif
+SASUMKERNEL    = sasum_thunderx2t99.c
+DASUMKERNEL    = dasum_thunderx2t99.c
+CASUMKERNEL    = casum_thunderx2t99.c
+ZASUMKERNEL    = zasum_thunderx2t99.c
+
+SCOPYKERNEL    = copy_thunderx2t99.c
+DCOPYKERNEL    = copy_thunderx2t99.c
+CCOPYKERNEL    = copy_thunderx2t99.c
+ZCOPYKERNEL    = copy_thunderx2t99.c
+
+SSWAPKERNEL    = swap_thunderx2t99.S
+DSWAPKERNEL    = swap_thunderx2t99.S
+CSWAPKERNEL    = swap_thunderx2t99.S
+ZSWAPKERNEL    = swap_thunderx2t99.S
+
+ISAMAXKERNEL   = iamax_thunderx2t99.c
+IDAMAXKERNEL   = iamax_thunderx2t99.c
+ICAMAXKERNEL   = izamax_thunderx2t99.c
+IZAMAXKERNEL   = izamax_thunderx2t99.c
+
+SNRM2KERNEL    = scnrm2_thunderx2t99.c
+DNRM2KERNEL    = dznrm2_thunderx2t99.c
+CNRM2KERNEL    = scnrm2_thunderx2t99.c
+ZNRM2KERNEL    = dznrm2_thunderx2t99.c
+
+DDOTKERNEL     = dot.c
+SDOTKERNEL     = dot.c
+CDOTKERNEL     = zdot_thunderx2t99.c
+ZDOTKERNEL     = zdot_thunderx2t99.c
 DSDOTKERNEL    = dot.S
 
 DGEMM_BETA     = dgemm_beta.S
@@ -128,10 +118,10 @@ SGEMM_BETA     = sgemm_beta.S
 SGEMMKERNEL    =  sgemm_kernel_sve_v2x$(SGEMM_UNROLL_N).S
 STRMMKERNEL    =  strmm_kernel_sve_v1x$(SGEMM_UNROLL_N).S
 
-SGEMMINCOPY    =  sgemm_ncopy_sve_v1.c
-SGEMMITCOPY    =  sgemm_tcopy_sve_v1.c
-SGEMMONCOPY    =  sgemm_ncopy_$(DGEMM_UNROLL_N).S
-SGEMMOTCOPY    =  sgemm_tcopy_$(DGEMM_UNROLL_N).S
+SGEMMINCOPY    =  gemm_ncopy_sve_v1x$(SGEMM_UNROLL_N).c
+SGEMMITCOPY    =  gemm_tcopy_sve_v1x$(SGEMM_UNROLL_N).c
+SGEMMONCOPY    =  sgemm_ncopy_$(SGEMM_UNROLL_N).S
+SGEMMOTCOPY    =  sgemm_tcopy_$(SGEMM_UNROLL_N).S
 
 SGEMMINCOPYOBJ =  sgemm_incopy$(TSUFFIX).$(SUFFIX)
 SGEMMITCOPYOBJ =  sgemm_itcopy$(TSUFFIX).$(SUFFIX)
@@ -149,8 +139,8 @@ SSYMMLCOPY_M    =  symm_lcopy_sve.c
 DGEMMKERNEL    =  dgemm_kernel_sve_v2x$(DGEMM_UNROLL_N).S
 DTRMMKERNEL    =  dtrmm_kernel_sve_v1x$(DGEMM_UNROLL_N).S
 
-DGEMMINCOPY    =  dgemm_ncopy_sve_v1.c
-DGEMMITCOPY    =  dgemm_tcopy_sve_v1.c
+DGEMMINCOPY    =  gemm_ncopy_sve_v1x$(DGEMM_UNROLL_N).c
+DGEMMITCOPY    =  gemm_tcopy_sve_v1x$(DGEMM_UNROLL_N).c
 DGEMMONCOPY    =  dgemm_ncopy_$(DGEMM_UNROLL_N).S
 DGEMMOTCOPY    =  dgemm_tcopy_$(DGEMM_UNROLL_N).S
 
@@ -170,8 +160,8 @@ DSYMMLCOPY_M    =  symm_lcopy_sve.c
 CGEMMKERNEL    =  cgemm_kernel_sve_v1x$(ZGEMM_UNROLL_N).S
 CTRMMKERNEL    =  ctrmm_kernel_sve_v1x$(ZGEMM_UNROLL_N).S
 
-CGEMMINCOPY    =  cgemm_ncopy_sve_v1.c
-CGEMMITCOPY    =  cgemm_tcopy_sve_v1.c
+CGEMMINCOPY    =  gemm_ncopy_complex_sve_v1x$(ZGEMM_UNROLL_N).c
+CGEMMITCOPY    =  gemm_tcopy_complex_sve_v1x$(ZGEMM_UNROLL_N).c
 CGEMMONCOPY    =  ../generic/zgemm_ncopy_$(ZGEMM_UNROLL_N).c
 CGEMMOTCOPY    =  ../generic/zgemm_tcopy_$(ZGEMM_UNROLL_N).c
 
@@ -194,8 +184,8 @@ CSYMMLCOPY_M    =  zsymm_lcopy_sve.c
 ZGEMMKERNEL    =  zgemm_kernel_sve_v1x$(ZGEMM_UNROLL_N).S
 ZTRMMKERNEL    =  ztrmm_kernel_sve_v1x$(ZGEMM_UNROLL_N).S
 
-ZGEMMINCOPY    =  zgemm_ncopy_sve_v1.c
-ZGEMMITCOPY    =  zgemm_tcopy_sve_v1.c
+ZGEMMINCOPY    =  gemm_ncopy_complex_sve_v1x$(ZGEMM_UNROLL_N).c
+ZGEMMITCOPY    =  gemm_tcopy_complex_sve_v1x$(ZGEMM_UNROLL_N).c
 ZGEMMONCOPY    =  ../generic/zgemm_ncopy_$(ZGEMM_UNROLL_N).c
 ZGEMMOTCOPY    =  ../generic/zgemm_tcopy_$(ZGEMM_UNROLL_N).c
 

kernel/arm64/KERNEL.NEOVERSEV1

@@ -1,189 +1 @@
-SAMINKERNEL  = ../arm/amin.c
-DAMINKERNEL  = ../arm/amin.c
-CAMINKERNEL  = ../arm/zamin.c
-ZAMINKERNEL  = ../arm/zamin.c
-
-SMAXKERNEL   = ../arm/max.c
-DMAXKERNEL   = ../arm/max.c
-
-SMINKERNEL   = ../arm/min.c
-DMINKERNEL   = ../arm/min.c
-
-ISAMINKERNEL = ../arm/iamin.c
-IDAMINKERNEL = ../arm/iamin.c
-ICAMINKERNEL = ../arm/izamin.c
-IZAMINKERNEL = ../arm/izamin.c
-
-ISMAXKERNEL  = ../arm/imax.c
-IDMAXKERNEL  = ../arm/imax.c
-
-ISMINKERNEL  = ../arm/imin.c
-IDMINKERNEL  = ../arm/imin.c
-
-STRSMKERNEL_LN	=  ../generic/trsm_kernel_LN.c
-STRSMKERNEL_LT	=  ../generic/trsm_kernel_LT.c
-STRSMKERNEL_RN	=  ../generic/trsm_kernel_RN.c
-STRSMKERNEL_RT	=  ../generic/trsm_kernel_RT.c
-
-DTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
-DTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
-DTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
-DTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
-
-CTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
-CTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
-CTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
-CTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
-
-ZTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
-ZTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
-ZTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
-ZTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
-
-SAMAXKERNEL  = amax.S
-DAMAXKERNEL  = amax.S
-CAMAXKERNEL  = zamax.S
-ZAMAXKERNEL  = zamax.S
-
-SAXPYKERNEL  = axpy.S
-DAXPYKERNEL  = daxpy_thunderx2t99.S
-CAXPYKERNEL  = zaxpy.S
-ZAXPYKERNEL  = zaxpy.S
-
-SROTKERNEL   = rot.S
-DROTKERNEL   = rot.S
-CROTKERNEL   = zrot.S
-ZROTKERNEL   = zrot.S
-
-SSCALKERNEL  = scal.S
-DSCALKERNEL  = scal.S
-CSCALKERNEL  = zscal.S
-ZSCALKERNEL  = zscal.S
-
-SGEMVNKERNEL = gemv_n.S
-DGEMVNKERNEL = gemv_n.S
-CGEMVNKERNEL = zgemv_n.S
-ZGEMVNKERNEL = zgemv_n.S
-
-SGEMVTKERNEL = gemv_t.S
-DGEMVTKERNEL = gemv_t.S
-CGEMVTKERNEL = zgemv_t.S
-ZGEMVTKERNEL = zgemv_t.S
-
-
-SASUMKERNEL    = sasum_thunderx2t99.c
-DASUMKERNEL    = dasum_thunderx2t99.c
-CASUMKERNEL    = casum_thunderx2t99.c
-ZASUMKERNEL    = zasum_thunderx2t99.c
-
-SCOPYKERNEL    = copy_thunderx2t99.c
-DCOPYKERNEL    = copy_thunderx2t99.c
-CCOPYKERNEL    = copy_thunderx2t99.c
-ZCOPYKERNEL    = copy_thunderx2t99.c
-
-SSWAPKERNEL    = swap_thunderx2t99.S
-DSWAPKERNEL    = swap_thunderx2t99.S
-CSWAPKERNEL    = swap_thunderx2t99.S
-ZSWAPKERNEL    = swap_thunderx2t99.S
-
-ISAMAXKERNEL   = iamax_thunderx2t99.c
-IDAMAXKERNEL   = iamax_thunderx2t99.c
-ICAMAXKERNEL   = izamax_thunderx2t99.c
-IZAMAXKERNEL   = izamax_thunderx2t99.c
-
-SNRM2KERNEL    = scnrm2_thunderx2t99.c
-DNRM2KERNEL    = dznrm2_thunderx2t99.c
-CNRM2KERNEL    = scnrm2_thunderx2t99.c
-ZNRM2KERNEL    = dznrm2_thunderx2t99.c
-
-DDOTKERNEL     = dot.c
-SDOTKERNEL     = dot.c
-CDOTKERNEL     = zdot_thunderx2t99.c
-ZDOTKERNEL     = zdot_thunderx2t99.c
-DSDOTKERNEL    = dot.S
-
-DGEMM_BETA     = dgemm_beta.S
-SGEMM_BETA     = sgemm_beta.S
-
-SGEMMKERNEL    =  sgemm_kernel_$(SGEMM_UNROLL_M)x$(SGEMM_UNROLL_N).S
-STRMMKERNEL    =  strmm_kernel_$(SGEMM_UNROLL_M)x$(SGEMM_UNROLL_N).S
-ifneq ($(SGEMM_UNROLL_M), $(SGEMM_UNROLL_N))
-ifeq ($(SGEMM_UNROLL_M), 16)
-SGEMMITCOPY    =  sgemm_tcopy_$(SGEMM_UNROLL_M).S
-else
-SGEMMITCOPY    =  ../generic/gemm_tcopy_$(SGEMM_UNROLL_M).c
-endif
-ifeq ($(SGEMM_UNROLL_M), 4)
-SGEMMINCOPY    =  sgemm_ncopy_$(SGEMM_UNROLL_M).S
-else
-SGEMMINCOPY    =  ../generic/gemm_ncopy_$(SGEMM_UNROLL_M).c
-endif
-SGEMMINCOPYOBJ =  sgemm_incopy$(TSUFFIX).$(SUFFIX)
-SGEMMITCOPYOBJ =  sgemm_itcopy$(TSUFFIX).$(SUFFIX)
-endif
-ifeq ($(SGEMM_UNROLL_N), 16)
-SGEMMOTCOPY    =  sgemm_tcopy_$(SGEMM_UNROLL_N).S
-else
-SGEMMOTCOPY    =  ../generic/gemm_tcopy_$(SGEMM_UNROLL_N).c
-endif
-ifeq ($(SGEMM_UNROLL_N), 4)
-SGEMMONCOPY    =  sgemm_ncopy_$(SGEMM_UNROLL_N).S
-else
-SGEMMONCOPY    =  ../generic/gemm_ncopy_$(SGEMM_UNROLL_N).c
-endif
-SGEMMONCOPYOBJ =  sgemm_oncopy$(TSUFFIX).$(SUFFIX)
-SGEMMOTCOPYOBJ =  sgemm_otcopy$(TSUFFIX).$(SUFFIX)
-
-DGEMMKERNEL    =  dgemm_kernel_$(DGEMM_UNROLL_M)x$(DGEMM_UNROLL_N).S
-DTRMMKERNEL    =  dtrmm_kernel_$(DGEMM_UNROLL_M)x$(DGEMM_UNROLL_N).S
-
-ifneq ($(DGEMM_UNROLL_M), $(DGEMM_UNROLL_N))
-
-ifeq ($(DGEMM_UNROLL_M), 8)
-DGEMMINCOPY    =  dgemm_ncopy_$(DGEMM_UNROLL_M).S
-DGEMMITCOPY    =  dgemm_tcopy_$(DGEMM_UNROLL_M).S
-else
-DGEMMINCOPY    =  ../generic/gemm_ncopy_$(DGEMM_UNROLL_M).c
-DGEMMITCOPY    =  ../generic/gemm_tcopy_$(DGEMM_UNROLL_M).c
-endif
-
-DGEMMINCOPYOBJ =  dgemm_incopy$(TSUFFIX).$(SUFFIX)
-DGEMMITCOPYOBJ =  dgemm_itcopy$(TSUFFIX).$(SUFFIX)
-endif
-
-ifeq ($(DGEMM_UNROLL_N), 4)
-DGEMMONCOPY    =  dgemm_ncopy_$(DGEMM_UNROLL_N).S
-DGEMMOTCOPY    =  dgemm_tcopy_$(DGEMM_UNROLL_N).S
-else
-DGEMMONCOPY    =  ../generic/gemm_ncopy_$(DGEMM_UNROLL_N).c
-DGEMMOTCOPY    =  ../generic/gemm_tcopy_$(DGEMM_UNROLL_N).c
-endif
-
-DGEMMONCOPYOBJ =  dgemm_oncopy$(TSUFFIX).$(SUFFIX)
-DGEMMOTCOPYOBJ =  dgemm_otcopy$(TSUFFIX).$(SUFFIX)
-
-CGEMMKERNEL    =  cgemm_kernel_$(CGEMM_UNROLL_M)x$(CGEMM_UNROLL_N).S
-CTRMMKERNEL    =  ctrmm_kernel_$(CGEMM_UNROLL_M)x$(CGEMM_UNROLL_N).S
-ifneq ($(CGEMM_UNROLL_M), $(CGEMM_UNROLL_N))
-CGEMMINCOPY    =  ../generic/zgemm_ncopy_$(CGEMM_UNROLL_M).c
-CGEMMITCOPY    =  ../generic/zgemm_tcopy_$(CGEMM_UNROLL_M).c
-CGEMMINCOPYOBJ =  cgemm_incopy$(TSUFFIX).$(SUFFIX)
-CGEMMITCOPYOBJ =  cgemm_itcopy$(TSUFFIX).$(SUFFIX)
-endif
-CGEMMONCOPY    =  ../generic/zgemm_ncopy_$(CGEMM_UNROLL_N).c
-CGEMMOTCOPY    =  ../generic/zgemm_tcopy_$(CGEMM_UNROLL_N).c
-CGEMMONCOPYOBJ =  cgemm_oncopy$(TSUFFIX).$(SUFFIX)
-CGEMMOTCOPYOBJ =  cgemm_otcopy$(TSUFFIX).$(SUFFIX)
-
-ZGEMMKERNEL    =  zgemm_kernel_$(ZGEMM_UNROLL_M)x$(ZGEMM_UNROLL_N).S
-ZTRMMKERNEL    =  ztrmm_kernel_$(ZGEMM_UNROLL_M)x$(ZGEMM_UNROLL_N).S
-ifneq ($(ZGEMM_UNROLL_M), $(ZGEMM_UNROLL_N))
-ZGEMMINCOPY    =  ../generic/zgemm_ncopy_$(ZGEMM_UNROLL_M).c
-ZGEMMITCOPY    =  ../generic/zgemm_tcopy_$(ZGEMM_UNROLL_M).c
-ZGEMMINCOPYOBJ =  zgemm_incopy$(TSUFFIX).$(SUFFIX)
-ZGEMMITCOPYOBJ =  zgemm_itcopy$(TSUFFIX).$(SUFFIX)
-endif
-ZGEMMONCOPY    =  ../generic/zgemm_ncopy_$(ZGEMM_UNROLL_N).c
-ZGEMMOTCOPY    =  ../generic/zgemm_tcopy_$(ZGEMM_UNROLL_N).c
-ZGEMMONCOPYOBJ =  zgemm_oncopy$(TSUFFIX).$(SUFFIX)
-ZGEMMOTCOPYOBJ =  zgemm_otcopy$(TSUFFIX).$(SUFFIX)
+include $(KERNELDIR)/KERNEL.ARMV8SVE

kernel/arm64/cgemm_kernel_8x4.S

@@ -49,7 +49,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define pCRow3		x15
 #define pA		x16
 #define alphaR		w17
-#define alphaI		w18
+#define alphaI		w19
 
 #define alpha0_R	s10
 #define alphaV0_R	v10.s[0]

kernel/arm64/cgemm_kernel_8x4_thunderx2t99.S

@@ -49,7 +49,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define pCRow3		x15
 #define pA		x16
 #define alphaR		w17
-#define alphaI		w18
+#define alphaI		w19
 
 #define alpha0_R	s10
 #define alphaV0_R	v10.s[0]

kernel/arm64/cgemm_kernel_sve_v1x4.S

@@ -240,7 +240,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
     add pB, pB, 32
 
-	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE+64]
 .endm
 
 .macro KERNELv1x4_M1
@@ -276,9 +275,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
     ld1rw  z15.s, p0/z,  [pB, 28]
 
     add pB, pB, 32
-	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
-
-	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE+64]
 .endm
 
 .macro KERNELv1x4_M2
@@ -313,11 +309,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	OP_ri	z23.s, p1/m, z2.s, z15.s
     ld1rw  z15.s, p0/z,  [pB, 28]
 
-	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
-
     add pB, pB, 32
-
-	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE+64]
 .endm
 
 .macro KERNELv1x4_E
@@ -341,10 +333,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	OP_ii	z22.s, p1/m, z3.s, z15.s
 	OP_ri	z23.s, p1/m, z2.s, z15.s
 
-	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
-
-	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE+64]
-
 .endm
 
 .macro KERNELv1x4_SUB
@@ -383,13 +371,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	OP_ii	z22.s, p1/m, z1.s, z15.s
 	OP_ri	z23.s, p1/m, z0.s, z15.s
 
-	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
-	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
 .endm
 
 .macro SAVEv1x4
-	prfm	PLDL2KEEP, [pCRow0, #C_PRE_SIZE]
-
 	ld2w	{z24.s, z25.s}, p1/z, [pCRow0]
 	fmla	z24.s, p1/m, z16.s, alphaz_R
 	fmls	z24.s, p1/m, z17.s, alphaz_I
@@ -407,8 +391,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	st2w 	{z26.s, z27.s}, p1, [pCRow1]
 
 	add	pCRow1, pCRow1, lanes, lsl #3
-	prfm	PLDL2KEEP, [pCRow1, #C_PRE_SIZE]
-
 	ld2w	{z28.s, z29.s}, p1/z, [pCRow2]
 	fmla	z28.s, p1/m, z20.s, alphaz_R
 	fmls	z28.s, p1/m, z21.s, alphaz_I
@@ -425,12 +407,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	fmla	z31.s, p1/m, z23.s, alphaz_R
 	st2w 	{z30.s, z31.s}, p1, [pCRow3]
 
-	prfm	PLDL2KEEP, [pCRow3, #C_PRE_SIZE]
-
 	add	pCRow3, pCRow3, lanes, lsl #3	// pC = pC + lanes  * 2 *4
 
-	prfm	PLDL2KEEP, [pCRow3, #C_PRE_SIZE]
-
 .endm
 
 /******************************************************************************/
@@ -466,8 +444,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 .endm
 
 .macro SAVEv1x2
-	prfm	PLDL2KEEP, [pCRow0, #C_PRE_SIZE]
-
 	ld2w	{z24.s, z25.s}, p1/z, [pCRow0]
 	fmla	z24.s, p1/m, z16.s, alphaz_R
 	fmls	z24.s, p1/m, z17.s, alphaz_I
@@ -485,10 +461,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	st2w 	{z26.s, z27.s}, p1, [pCRow1]
 
 	add	pCRow1, pCRow1, lanes, lsl #3
-	prfm	PLDL2KEEP, [pCRow1, #C_PRE_SIZE]
-
-	prfm	PLDL2KEEP, [pCRow2, #C_PRE_SIZE]
-
 .endm
 
 /******************************************************************************/
@@ -516,8 +488,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 .endm
 
 .macro SAVEv1x1
-	prfm	PLDL2KEEP, [pCRow0, #C_PRE_SIZE]
-
 	ld2w	{z24.s, z25.s}, p1/z, [pCRow0]
 	fmla	z24.s, p1/m, z16.s, alphaz_R
 	fmls	z24.s, p1/m, z17.s, alphaz_I
@@ -527,8 +497,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 	add	pCRow0, pCRow0, lanes, lsl #3	// pC = pC + lanes  * 2 *4
 
-	prfm	PLDL2KEEP, [pCRow3, #C_PRE_SIZE]
-
 .endm
 
 /******************************************************************************/
@@ -553,9 +521,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	stp	x26, x27, [sp, #(9 * 16)]
 	str	x28, [sp, #(10 * 16)]
 
-	prfm	PLDL1KEEP, [origPB]
-	prfm	PLDL1KEEP, [origPA]
-
 	fmov	alphaR, s0
 	dup	    alphaz_R, alphaR
 	fmov	alphaI, s1
@@ -676,10 +641,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	bne	.Lcgemm_kernel_L4_Mv1_46
 
 .Lcgemm_kernel_L4_Mv1_100:
-	prfm	PLDL1KEEP, [pA]
-	prfm	PLDL1KEEP, [pA, #64]
-	prfm	PLDL1KEEP, [origPB]
-
 	SAVEv1x4
 
 .Lcgemm_kernel_L4_Mv1_END:

kernel/arm64/cgemm_ncopy_sve_v1.c

@@ -1,5 +1,6 @@
 /*********************************************************************/
 /* Copyright 2009, 2010 The University of Texas at Austin.           */
+/* Copyright 2023 The OpenBLAS Project                               */
 /* All rights reserved.                                              */
 /*                                                                   */
 /* Redistribution and use in source and binary forms, with or        */
@@ -52,7 +53,7 @@ int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b){
     boffset = b;
 
     j = 0;
-    svbool_t pg = svwhilelt_b32(j, n);
+    svbool_t pg = svwhilelt_b32((uint64_t)j, (uint64_t)n);
     uint32_t active = svcntp_b32(svptrue_b32(), pg);
     do {
 
@@ -69,7 +70,7 @@ int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b){
         aoffset += active * lda * 2;
 
         j += svcntw();
-        pg = svwhilelt_b32(j, n);
+        pg = svwhilelt_b32((uint64_t)j, (uint64_t)n);
         active = svcntp_b32(svptrue_b32(), pg);
 
 

kernel/arm64/cgemm_tcopy_sve_v1.c

@@ -1,5 +1,6 @@
 /*********************************************************************/
 /* Copyright 2009, 2010 The University of Texas at Austin.           */
+/* Copyright 2023 The OpenBLAS Project                               */
 /* All rights reserved.                                              */
 /*                                                                   */
 /* Redistribution and use in source and binary forms, with or        */
@@ -50,7 +51,7 @@ int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b){
     boffset = b;
 
     j = 0;
-    svbool_t pg = svwhilelt_b32(j, n);
+    svbool_t pg = svwhilelt_b32((uint64_t)j, (uint64_t)n);
     uint32_t active = svcntp_b32(svptrue_b32(), pg);
     do {
 
@@ -66,7 +67,7 @@ int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b){
         aoffset += active * 2;
 
         j += svcntw();
-        pg = svwhilelt_b32(j, n);
+        pg = svwhilelt_b32((uint64_t)j, (uint64_t)n);
         active = svcntp_b32(svptrue_b32(), pg);
 
     } while (svptest_any(svptrue_b32(), pg));

kernel/arm64/ctrmm_kernel_8x4.S

@@ -49,10 +49,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define pCRow3		x15
 #define pA		x16
 #define alphaR		w17
-#define alphaI		w18
-#define temp		x19
-#define tempOffset	x20
-#define tempK		x21
+#define alphaI		w19
+#define temp		x20
+#define tempOffset	x21
+#define tempK		x22
 
 #define alpha0_R	s10
 #define alphaV0_R	v10.s[0]

kernel/arm64/dgemm_ncopy_sve_v1.c

@@ -1,79 +0,0 @@
-/*********************************************************************/
-/* Copyright 2009, 2010 The University of Texas at Austin.           */
-/* All rights reserved.                                              */
-/*                                                                   */
-/* Redistribution and use in source and binary forms, with or        */
-/* without modification, are permitted provided that the following   */
-/* conditions are met:                                               */
-/*                                                                   */
-/*   1. Redistributions of source code must retain the above         */
-/*      copyright notice, this list of conditions and the following  */
-/*      disclaimer.                                                  */
-/*                                                                   */
-/*   2. Redistributions in binary form must reproduce the above      */
-/*      copyright notice, this list of conditions and the following  */
-/*      disclaimer in the documentation and/or other materials       */
-/*      provided with the distribution.                              */
-/*                                                                   */
-/*    THIS  SOFTWARE IS PROVIDED  BY THE  UNIVERSITY OF  TEXAS AT    */
-/*    AUSTIN  ``AS IS''  AND ANY  EXPRESS OR  IMPLIED WARRANTIES,    */
-/*    INCLUDING, BUT  NOT LIMITED  TO, THE IMPLIED  WARRANTIES OF    */
-/*    MERCHANTABILITY  AND FITNESS FOR  A PARTICULAR  PURPOSE ARE    */
-/*    DISCLAIMED.  IN  NO EVENT SHALL THE UNIVERSITY  OF TEXAS AT    */
-/*    AUSTIN OR CONTRIBUTORS BE  LIABLE FOR ANY DIRECT, INDIRECT,    */
-/*    INCIDENTAL,  SPECIAL, EXEMPLARY,  OR  CONSEQUENTIAL DAMAGES    */
-/*    (INCLUDING, BUT  NOT LIMITED TO,  PROCUREMENT OF SUBSTITUTE    */
-/*    GOODS  OR  SERVICES; LOSS  OF  USE,  DATA,  OR PROFITS;  OR    */
-/*    BUSINESS INTERRUPTION) HOWEVER CAUSED  AND ON ANY THEORY OF    */
-/*    LIABILITY, WHETHER  IN CONTRACT, STRICT  LIABILITY, OR TORT    */
-/*    (INCLUDING NEGLIGENCE OR OTHERWISE)  ARISING IN ANY WAY OUT    */
-/*    OF  THE  USE OF  THIS  SOFTWARE,  EVEN  IF ADVISED  OF  THE    */
-/*    POSSIBILITY OF SUCH DAMAGE.                                    */
-/*                                                                   */
-/* The views and conclusions contained in the software and           */
-/* documentation are those of the authors and should not be          */
-/* interpreted as representing official policies, either expressed   */
-/* or implied, of The University of Texas at Austin.                 */
-/*********************************************************************/
-
-#include <stdio.h>
-#include "common.h"
-#include <arm_sve.h>
-
-// TODO: write in assembly with proper unrolling of inner loop
-int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b){
-
-    BLASLONG j;
-    IFLOAT *aoffset, *aoffset1, *boffset;
-
-    svint64_t lda_vec = svindex_s64(0LL, lda);
-    uint64_t sve_size = svcntd();
-
-    aoffset = a;
-    boffset = b;
-
-    j = 0;
-    svbool_t pg = svwhilelt_b64(j, n);
-    uint64_t active = svcntp_b64(svptrue_b64(), pg);
-    do {
-
-        aoffset1 = aoffset;
-
-        uint64_t i_cnt = m;
-        while (i_cnt--) {
-            svfloat64_t a_vec = svld1_gather_index(pg, (double *) aoffset1, lda_vec);
-            svst1_f64(pg, (double *) boffset, a_vec);
-            aoffset1++;
-            boffset += active;
-        }
-        aoffset += sve_size * lda;
-
-        j += svcntd();
-        pg = svwhilelt_b64(j, n);
-        active = svcntp_b64(svptrue_b64(), pg);
-
-
-    } while (svptest_any(svptrue_b64(), pg));
-
-    return 0;
-}

kernel/arm64/dgemm_tcopy_sve_v1.c

@@ -1,77 +0,0 @@
-/*********************************************************************/
-/* Copyright 2009, 2010 The University of Texas at Austin.           */
-/* All rights reserved.                                              */
-/*                                                                   */
-/* Redistribution and use in source and binary forms, with or        */
-/* without modification, are permitted provided that the following   */
-/* conditions are met:                                               */
-/*                                                                   */
-/*   1. Redistributions of source code must retain the above         */
-/*      copyright notice, this list of conditions and the following  */
-/*      disclaimer.                                                  */
-/*                                                                   */
-/*   2. Redistributions in binary form must reproduce the above      */
-/*      copyright notice, this list of conditions and the following  */
-/*      disclaimer in the documentation and/or other materials       */
-/*      provided with the distribution.                              */
-/*                                                                   */
-/*    THIS  SOFTWARE IS PROVIDED  BY THE  UNIVERSITY OF  TEXAS AT    */
-/*    AUSTIN  ``AS IS''  AND ANY  EXPRESS OR  IMPLIED WARRANTIES,    */
-/*    INCLUDING, BUT  NOT LIMITED  TO, THE IMPLIED  WARRANTIES OF    */
-/*    MERCHANTABILITY  AND FITNESS FOR  A PARTICULAR  PURPOSE ARE    */
-/*    DISCLAIMED.  IN  NO EVENT SHALL THE UNIVERSITY  OF TEXAS AT    */
-/*    AUSTIN OR CONTRIBUTORS BE  LIABLE FOR ANY DIRECT, INDIRECT,    */
-/*    INCIDENTAL,  SPECIAL, EXEMPLARY,  OR  CONSEQUENTIAL DAMAGES    */
-/*    (INCLUDING, BUT  NOT LIMITED TO,  PROCUREMENT OF SUBSTITUTE    */
-/*    GOODS  OR  SERVICES; LOSS  OF  USE,  DATA,  OR PROFITS;  OR    */
-/*    BUSINESS INTERRUPTION) HOWEVER CAUSED  AND ON ANY THEORY OF    */
-/*    LIABILITY, WHETHER  IN CONTRACT, STRICT  LIABILITY, OR TORT    */
-/*    (INCLUDING NEGLIGENCE OR OTHERWISE)  ARISING IN ANY WAY OUT    */
-/*    OF  THE  USE OF  THIS  SOFTWARE,  EVEN  IF ADVISED  OF  THE    */
-/*    POSSIBILITY OF SUCH DAMAGE.                                    */
-/*                                                                   */
-/* The views and conclusions contained in the software and           */
-/* documentation are those of the authors and should not be          */
-/* interpreted as representing official policies, either expressed   */
-/* or implied, of The University of Texas at Austin.                 */
-/*********************************************************************/
-
-#include <stdio.h>
-#include "common.h"
-#include <arm_sve.h>
-
-// TODO: write in assembly with proper unrolling of inner loop
-int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b){
-
-    BLASLONG j;
-    IFLOAT *aoffset, *aoffset1, *boffset;
-
-    uint64_t sve_size = svcntd();
-
-    aoffset = a;
-    boffset = b;
-
-    j = 0;
-    svbool_t pg = svwhilelt_b64(j, n);
-    uint64_t active = svcntp_b64(svptrue_b64(), pg);
-    do {
-
-        aoffset1 = aoffset;
-
-        uint64_t i_cnt = m;
-        while (i_cnt--) {
-            svfloat64_t a_vec = svld1(pg, (double *)aoffset1);
-            svst1_f64(pg, (double *) boffset, a_vec);
-            aoffset1 += lda;
-            boffset += active;
-        }
-        aoffset += sve_size;
-
-        j += svcntd();
-        pg = svwhilelt_b64(j, n);
-        active = svcntp_b64(svptrue_b64(), pg);
-
-    } while (svptest_any(svptrue_b64(), pg));
-
-    return 0;
-}

kernel/arm64/dot_kernel_sve.c

@@ -50,8 +50,8 @@ static FLOAT dot_kernel_sve(BLASLONG n, FLOAT *x, FLOAT *y) {
         BLASLONG sve_width = SVE_WIDTH;
 
         for (BLASLONG i = 0; i < n; i += sve_width * 2) {
-                svbool_t pg_a = SVE_WHILELT(i, n);
-                svbool_t pg_b = SVE_WHILELT(i + sve_width, n);
+                svbool_t pg_a = SVE_WHILELT((uint64_t)i, (uint64_t)n);
+                svbool_t pg_b = SVE_WHILELT((uint64_t)(i + sve_width), (uint64_t)n);
 
                 SVE_TYPE x_vec_a = svld1(pg_a, &x[i]);
                 SVE_TYPE y_vec_a = svld1(pg_a, &y[i]);

kernel/arm64/dznrm2_thunderx2t99.c

@@ -27,7 +27,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 
 #include "common.h"
-
+#include <float.h>
 #include <arm_neon.h>
 
 #if defined(SMP)
@@ -404,7 +404,8 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
 #else
 	nrm2_compute(n, x, inc_x, &ssq, &scale);
 #endif
-	if (fabs(scale) <1.e-300) return 0.;
+	volatile FLOAT sca = fabs(scale);
+	if (sca < DBL_MIN) return 0.;
 	ssq = sqrt(ssq) * scale;
 
 	return ssq;

kernel/arm64/gemm_ncopy_complex_sve_v1x4.c

@@ -0,0 +1,121 @@
+/***************************************************************************
+Copyright (c) 2023, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A00 PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include <stdint.h>
+#include <stdio.h>
+#include <arm_sve.h>
+
+#include "common.h"
+
+#ifdef DOUBLE
+#define COUNT "cntd"
+#define SV_TYPE svfloat64_t
+#define SV_INDEX svuint64_t
+#define SV_INDEXER svindex_u64
+#define SV_TRUE svptrue_b64
+#define SV_WHILE svwhilelt_b64
+#else
+#define COUNT "cntw"
+#define SV_TYPE svfloat32_t
+#define SV_INDEX svuint32_t
+#define SV_INDEXER svindex_u32
+#define SV_TRUE svptrue_b32
+#define SV_WHILE svwhilelt_b32
+#endif
+
+#define INNER_COPY(pg, a_offset_inner, b_offset, lda, active)           \
+    a_vec_real = svld1_gather_index(pg, a_offset_inner, lda_vec);       \
+    a_vec_imag = svld1_gather_index(pg, a_offset_inner + 1, lda_vec);   \
+    svst2(pg, b_offset, svcreate2(a_vec_real, a_vec_imag));             \
+    a_offset_inner += 2;                                                \
+    b_offset += active * 2;
+
+int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b) {
+    uint64_t sve_size;
+    asm(COUNT" %[SIZE_]" : [SIZE_]  "=r" (sve_size) : : );
+
+    IFLOAT *a_offset, *a_offset_inner, *b_offset;
+    a_offset = a;
+    b_offset = b;
+
+    SV_INDEX lda_vec = SV_INDEXER(0LL, lda * 2);
+    SV_TYPE a_vec_real;
+    SV_TYPE a_vec_imag;
+    svbool_t pg_true = SV_TRUE();
+
+    BLASLONG single_vectors_n = n & -sve_size;
+    for (BLASLONG j = 0; j < single_vectors_n; j += sve_size) {
+        a_offset_inner = a_offset;
+
+        svbool_t pg = pg_true;
+        uint64_t active = sve_size;
+        uint64_t i_cnt = m >> 2;
+        while (i_cnt--) {
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+        }
+
+        if (m & 2) {
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+        }
+
+        if (m & 1) {
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+        }
+
+        a_offset += sve_size * lda * 2;
+    }
+
+    BLASLONG remaining_n = n - single_vectors_n;
+    if (remaining_n) {
+        a_offset_inner = a_offset;
+        svbool_t pg = SV_WHILE((uint64_t)0L, (uint64_t)remaining_n);
+        uint64_t active = remaining_n;
+        uint64_t i_cnt = m >> 2;
+        while (i_cnt--) {
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+        }
+
+        if (m & 2) {
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+        }
+
+        if (m & 1) {
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+        }
+    }
+
+    return 0;
+}
+

kernel/arm64/gemm_ncopy_sve_v1x8.c

@@ -0,0 +1,131 @@
+/***************************************************************************
+Copyright (c) 2023, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A00 PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include <stdint.h>
+#include <stdio.h>
+#include <arm_sve.h>
+
+#include "common.h"
+
+#ifdef DOUBLE
+#define COUNT "cntd"
+#define SV_TYPE svfloat64_t
+#define SV_INDEX svuint64_t
+#define SV_INDEXER svindex_u64
+#define SV_TRUE svptrue_b64
+#define SV_WHILE svwhilelt_b64
+#define SV_PREFETCH svprfd_gather_index
+#else
+#define COUNT "cntw"
+#define SV_TYPE svfloat32_t
+#define SV_INDEX svuint32_t
+#define SV_INDEXER svindex_u32
+#define SV_TRUE svptrue_b32
+#define SV_WHILE svwhilelt_b32
+#define SV_PREFETCH svprfw_gather_index
+#endif
+
+#define INNER_COPY(pg, a_offset_inner, b_offset, lda, active)   \
+    a_vec = svld1_gather_index(pg, a_offset_inner, lda_vec);    \
+    svst1(pg, b_offset, a_vec);                                 \
+    a_offset_inner++;                                           \
+    b_offset += active;
+
+int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b) {
+    uint64_t sve_size;
+    asm(COUNT" %[SIZE_]" : [SIZE_]  "=r" (sve_size) : : );
+
+    IFLOAT *a_offset, *a_offset_inner, *b_offset;
+    a_offset = a;
+    b_offset = b;
+
+    SV_INDEX lda_vec = SV_INDEXER(0LL, lda);
+    SV_TYPE a_vec;
+    svbool_t pg_true = SV_TRUE();
+
+    BLASLONG single_vectors_n = n & -sve_size;
+    for (BLASLONG j = 0; j < single_vectors_n; j += sve_size) {
+        a_offset_inner = a_offset;
+
+        svbool_t pg = pg_true;
+        uint64_t active = sve_size;
+        uint64_t i_cnt = m >> 3;
+        while (i_cnt--) {
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+        }
+
+        if (m & 4) {
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+        }
+
+        if (m & 2) {
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+        }
+
+        if (m & 1) {
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+        }
+
+        a_offset += sve_size * lda;
+    }
+
+    BLASLONG remaining_n = n - single_vectors_n;
+    if (remaining_n) {
+        a_offset_inner = a_offset;
+        svbool_t pg = SV_WHILE((uint64_t)0L, (uint64_t)remaining_n);
+        uint64_t active = remaining_n;
+        uint64_t i_cnt = m >> 2;
+        while (i_cnt--) {
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+        }
+
+        if (m & 2) {
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+        }
+
+        if (m & 1) {
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+        }
+    }
+
+    return 0;
+}

kernel/arm64/gemm_tcopy_complex_sve_v1x4.c

@@ -0,0 +1,115 @@
+/***************************************************************************
+Copyright (c) 2023, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A00 PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include <stdint.h>
+#include <stdio.h>
+#include <arm_sve.h>
+
+#include "common.h"
+
+#ifdef DOUBLE
+#define COUNT "cntd"
+#define SV_TYPE svfloat64x2_t
+#define SV_TRUE svptrue_b64
+#define SV_WHILE svwhilelt_b64
+#else
+#define COUNT "cntw"
+#define SV_TYPE svfloat32x2_t
+#define SV_TRUE svptrue_b32
+#define SV_WHILE svwhilelt_b32
+#endif
+
+#define INNER_COPY(pg, a_offset_inner, b_offset, lda, active)   \
+    a_vec = svld2(pg, a_offset_inner);                          \
+    svst2(pg, b_offset, a_vec);                                 \
+    a_offset_inner += lda * 2;                                  \
+    b_offset += active * 2;
+
+int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b){
+    uint64_t sve_size = svcntw();
+    asm(COUNT" %[SIZE_]" : [SIZE_]  "=r" (sve_size) : : );
+
+    IFLOAT *a_offset, *a_offset_inner, *b_offset;
+    a_offset = a;
+    b_offset = b;
+
+    SV_TYPE a_vec;
+    svbool_t pg_true = SV_TRUE();
+
+    BLASLONG single_vectors_n = n & -sve_size;
+    for (BLASLONG j = 0; j < single_vectors_n; j += sve_size) {
+        a_offset_inner = a_offset;
+
+        svbool_t pg = pg_true;
+        uint64_t active = sve_size;
+        uint64_t i_cnt = m >> 2;
+        while (i_cnt--) {
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+        }
+
+        if (m & 2) {
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+        }
+
+        if (m & 1) {
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+        }
+
+        a_offset += sve_size * 2;
+    }
+
+    BLASLONG remaining_n = n - single_vectors_n;
+    if (remaining_n) {
+        a_offset_inner = a_offset;
+        svbool_t pg = SV_WHILE((uint64_t)0L, (uint64_t)remaining_n);
+        uint64_t active = remaining_n;
+        uint64_t i_cnt = m >> 2;
+        while (i_cnt--) {
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+        }
+
+        if (m & 2) {
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+        }
+
+        if (m & 1) {
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+        }
+    }
+
+    return 0;
+}
+
+

kernel/arm64/gemm_tcopy_sve_v1x8.c

@@ -0,0 +1,125 @@
+/***************************************************************************
+Copyright (c) 2023, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A00 PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include <stdint.h>
+#include <stdio.h>
+#include <arm_sve.h>
+
+#include "common.h"
+
+#ifdef DOUBLE
+#define COUNT "cntd"
+#define SV_TYPE svfloat64_t
+#define SV_TRUE svptrue_b64
+#define SV_WHILE svwhilelt_b64
+#else
+#define COUNT "cntw"
+#define SV_TYPE svfloat32_t
+#define SV_TRUE svptrue_b32
+#define SV_WHILE svwhilelt_b32
+#endif
+
+#define INNER_COPY(pg, a_offset_inner, b_offset, lda, active)   \
+    a_vec = svld1(pg, a_offset_inner);                          \
+    svst1(pg, b_offset, a_vec);                                 \
+    a_offset_inner += lda;                                      \
+    b_offset += active;
+
+int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b){
+    uint64_t sve_size = svcntw();
+    asm(COUNT" %[SIZE_]" : [SIZE_]  "=r" (sve_size) : : );
+
+    IFLOAT *a_offset, *a_offset_inner, *b_offset;
+    a_offset = a;
+    b_offset = b;
+
+    SV_TYPE a_vec;
+    svbool_t pg_true = SV_TRUE();
+
+    BLASLONG single_vectors_n = n & -sve_size;
+    for (BLASLONG j = 0; j < single_vectors_n; j += sve_size) {
+        a_offset_inner = a_offset;
+
+        svbool_t pg = pg_true;
+        uint64_t active = sve_size;
+        uint64_t i_cnt = m >> 3;
+        while (i_cnt--) {
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+        }
+
+        if (m & 4) {
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+        }
+
+        if (m & 2) {
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+        }
+
+        if (m & 1) {
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+        }
+
+        a_offset += sve_size;
+    }
+
+    BLASLONG remaining_n = n - single_vectors_n;
+    if (remaining_n) {
+        a_offset_inner = a_offset;
+        svbool_t pg = SV_WHILE((uint64_t)0L, (uint64_t)remaining_n);
+        uint64_t active = remaining_n;
+        uint64_t i_cnt = m >> 2;
+        while (i_cnt--) {
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+        }
+
+        if (m & 2) {
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+        }
+
+        if (m & 1) {
+            INNER_COPY(pg, a_offset_inner, b_offset, lda, active);
+        }
+    }
+
+    return 0;
+}
+

kernel/arm64/sgemm_ncopy_sve_v1.c

@@ -1,78 +0,0 @@
-/*********************************************************************/
-/* Copyright 2009, 2010 The University of Texas at Austin.           */
-/* All rights reserved.                                              */
-/*                                                                   */
-/* Redistribution and use in source and binary forms, with or        */
-/* without modification, are permitted provided that the following   */
-/* conditions are met:                                               */
-/*                                                                   */
-/*   1. Redistributions of source code must retain the above         */
-/*      copyright notice, this list of conditions and the following  */
-/*      disclaimer.                                                  */
-/*                                                                   */
-/*   2. Redistributions in binary form must reproduce the above      */
-/*      copyright notice, this list of conditions and the following  */
-/*      disclaimer in the documentation and/or other materials       */
-/*      provided with the distribution.                              */
-/*                                                                   */
-/*    THIS  SOFTWARE IS PROVIDED  BY THE  UNIVERSITY OF  TEXAS AT    */
-/*    AUSTIN  ``AS IS''  AND ANY  EXPRESS OR  IMPLIED WARRANTIES,    */
-/*    INCLUDING, BUT  NOT LIMITED  TO, THE IMPLIED  WARRANTIES OF    */
-/*    MERCHANTABILITY  AND FITNESS FOR  A PARTICULAR  PURPOSE ARE    */
-/*    DISCLAIMED.  IN  NO EVENT SHALL THE UNIVERSITY  OF TEXAS AT    */
-/*    AUSTIN OR CONTRIBUTORS BE  LIABLE FOR ANY DIRECT, INDIRECT,    */
-/*    INCIDENTAL,  SPECIAL, EXEMPLARY,  OR  CONSEQUENTIAL DAMAGES    */
-/*    (INCLUDING, BUT  NOT LIMITED TO,  PROCUREMENT OF SUBSTITUTE    */
-/*    GOODS  OR  SERVICES; LOSS  OF  USE,  DATA,  OR PROFITS;  OR    */
-/*    BUSINESS INTERRUPTION) HOWEVER CAUSED  AND ON ANY THEORY OF    */
-/*    LIABILITY, WHETHER  IN CONTRACT, STRICT  LIABILITY, OR TORT    */
-/*    (INCLUDING NEGLIGENCE OR OTHERWISE)  ARISING IN ANY WAY OUT    */
-/*    OF  THE  USE OF  THIS  SOFTWARE,  EVEN  IF ADVISED  OF  THE    */
-/*    POSSIBILITY OF SUCH DAMAGE.                                    */
-/*                                                                   */
-/* The views and conclusions contained in the software and           */
-/* documentation are those of the authors and should not be          */
-/* interpreted as representing official policies, either expressed   */
-/* or implied, of The University of Texas at Austin.                 */
-/*********************************************************************/
-
-#include <stdio.h>
-#include "common.h"
-#include <arm_sve.h>
-
-// TODO: write in assembly with proper unrolling of inner loop
-int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b){
-
-    BLASLONG j;
-    IFLOAT *aoffset, *aoffset1, *boffset;
-
-    svint32_t lda_vec = svindex_s32(0LL, lda);
-    uint32_t sve_size = svcntw();
-
-    aoffset = a;
-    boffset = b;
-
-    j = 0;
-    svbool_t pg = svwhilelt_b32(j, n);
-    uint32_t active = svcntp_b32(svptrue_b32(), pg);
-    do {
-
-        aoffset1 = aoffset;
-
-        uint32_t i_cnt = m;
-        while (i_cnt--) {
-            svfloat32_t a_vec = svld1_gather_index(pg, (float *) aoffset1, lda_vec);
-            svst1_f32(pg, (float *) boffset, a_vec);
-            aoffset1++;
-            boffset += active;
-        }
-        aoffset += sve_size * lda;
-
-        j += svcntw();
-        pg = svwhilelt_b32(j, n);
-        active = svcntp_b32(svptrue_b32(), pg);
-
-    } while (svptest_any(svptrue_b32(), pg));
-
-    return 0;
-}

kernel/arm64/sgemm_tcopy_sve_v1.c

@@ -1,77 +0,0 @@
-/*********************************************************************/
-/* Copyright 2009, 2010 The University of Texas at Austin.           */
-/* All rights reserved.                                              */
-/*                                                                   */
-/* Redistribution and use in source and binary forms, with or        */
-/* without modification, are permitted provided that the following   */
-/* conditions are met:                                               */
-/*                                                                   */
-/*   1. Redistributions of source code must retain the above         */
-/*      copyright notice, this list of conditions and the following  */
-/*      disclaimer.                                                  */
-/*                                                                   */
-/*   2. Redistributions in binary form must reproduce the above      */
-/*      copyright notice, this list of conditions and the following  */
-/*      disclaimer in the documentation and/or other materials       */
-/*      provided with the distribution.                              */
-/*                                                                   */
-/*    THIS  SOFTWARE IS PROVIDED  BY THE  UNIVERSITY OF  TEXAS AT    */
-/*    AUSTIN  ``AS IS''  AND ANY  EXPRESS OR  IMPLIED WARRANTIES,    */
-/*    INCLUDING, BUT  NOT LIMITED  TO, THE IMPLIED  WARRANTIES OF    */
-/*    MERCHANTABILITY  AND FITNESS FOR  A PARTICULAR  PURPOSE ARE    */
-/*    DISCLAIMED.  IN  NO EVENT SHALL THE UNIVERSITY  OF TEXAS AT    */
-/*    AUSTIN OR CONTRIBUTORS BE  LIABLE FOR ANY DIRECT, INDIRECT,    */
-/*    INCIDENTAL,  SPECIAL, EXEMPLARY,  OR  CONSEQUENTIAL DAMAGES    */
-/*    (INCLUDING, BUT  NOT LIMITED TO,  PROCUREMENT OF SUBSTITUTE    */
-/*    GOODS  OR  SERVICES; LOSS  OF  USE,  DATA,  OR PROFITS;  OR    */
-/*    BUSINESS INTERRUPTION) HOWEVER CAUSED  AND ON ANY THEORY OF    */
-/*    LIABILITY, WHETHER  IN CONTRACT, STRICT  LIABILITY, OR TORT    */
-/*    (INCLUDING NEGLIGENCE OR OTHERWISE)  ARISING IN ANY WAY OUT    */
-/*    OF  THE  USE OF  THIS  SOFTWARE,  EVEN  IF ADVISED  OF  THE    */
-/*    POSSIBILITY OF SUCH DAMAGE.                                    */
-/*                                                                   */
-/* The views and conclusions contained in the software and           */
-/* documentation are those of the authors and should not be          */
-/* interpreted as representing official policies, either expressed   */
-/* or implied, of The University of Texas at Austin.                 */
-/*********************************************************************/
-
-#include <stdio.h>
-#include "common.h"
-#include <arm_sve.h>
-
-// TODO: write in assembly with proper unrolling of inner loop
-int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b){
-
-    BLASLONG j;
-    IFLOAT *aoffset, *aoffset1, *boffset;
-
-    uint32_t sve_size = svcntw();
-
-    aoffset = a;
-    boffset = b;
-
-    j = 0;
-    svbool_t pg = svwhilelt_b32(j, n);
-    uint32_t active = svcntp_b32(svptrue_b32(), pg);
-    do {
-
-        aoffset1 = aoffset;
-
-        uint32_t i_cnt = m;
-        while (i_cnt--) {
-            svfloat32_t a_vec = svld1(pg, (float *) aoffset1);
-            svst1_f32(pg, (float *) boffset, a_vec);
-            aoffset1 += lda;
-            boffset += active;
-        }
-        aoffset += sve_size;
-
-        j += svcntw();
-        pg = svwhilelt_b32(j, n);
-        active = svcntp_b32(svptrue_b32(), pg);
-
-    } while (svptest_any(svptrue_b32(), pg));
-
-    return 0;
-}

kernel/arm64/symm_lcopy_sve.c

@@ -1,5 +1,6 @@
 /*********************************************************************/
 /* Copyright 2009, 2010 The University of Texas at Austin.           */
+/* Copyright 2023 The OpenBLAS Project                               */
 /* All rights reserved.                                              */
 /*                                                                   */
 /* Redistribution and use in source and binary forms, with or        */
@@ -52,7 +53,7 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
   svint64_t one_vec = svdup_s64(1LL);
 
   int64_t j = 0;
-  svbool_t pg = svwhilelt_b64(j, n);
+  svbool_t pg = svwhilelt_b64((uint64_t)j, (uint64_t)n);
   int64_t active = svcntp_b64(svptrue_b64(), pg);
   svint64_t index_neg = svindex_s64(0LL, -1LL);
   svint64_t index = svindex_s64(0LL, 1LL);
@@ -86,7 +87,7 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
     posX += sve_size;
     posX_vec = svdup_s64(posX);
     j += sve_size;
-    pg = svwhilelt_b64(j, n);
+    pg = svwhilelt_b64((uint64_t)j, (uint64_t)n);
     active = svcntp_b64(svptrue_b64(), pg);
   } while (svptest_any(svptrue_b64(), pg));
 
@@ -99,7 +100,7 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
 
   int32_t N = n;
   int32_t j = 0;
-  svbool_t pg = svwhilelt_b32(j, N);
+  svbool_t pg = svwhilelt_b32((uint32_t)j, (uint32_t)N);
   int32_t active = svcntp_b32(svptrue_b32(), pg);
   svint32_t index_neg = svindex_s32(0, -1);
   svint32_t index = svindex_s32(0, 1);
@@ -133,7 +134,7 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
     posX += sve_size;
     posX_vec = svdup_s32(posX);
     j += sve_size;
-    pg = svwhilelt_b32(j, N);
+    pg = svwhilelt_b32((uint32_t)j, (uint32_t)N);
     active = svcntp_b32(svptrue_b32(), pg);
   } while (svptest_any(svptrue_b32(), pg));
 

kernel/arm64/symm_ucopy_sve.c

@@ -1,5 +1,6 @@
 /*********************************************************************/
 /* Copyright 2009, 2010 The University of Texas at Austin.           */
+/* Copyright 2023 The OpenBLAS Project                               */
 /* All rights reserved.                                              */
 /*                                                                   */
 /* Redistribution and use in source and binary forms, with or        */
@@ -52,7 +53,7 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
   svint64_t one_vec = svdup_s64(1LL);
 
   int64_t j = 0;
-  svbool_t pg = svwhilelt_b64(j, n);
+  svbool_t pg = svwhilelt_b64((uint64_t)j, (uint64_t)n);
   int64_t active = svcntp_b64(svptrue_b64(), pg);
   svint64_t index_neg = svindex_s64(0LL, -1LL);
   svint64_t index = svindex_s64(0LL, 1LL);
@@ -86,7 +87,7 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
     posX += sve_size;
     posX_vec = svdup_s64(posX);
     j += sve_size;
-    pg = svwhilelt_b64(j, n);
+    pg = svwhilelt_b64((uint64_t)j, (uint64_t)n);
     active = svcntp_b64(svptrue_b64(), pg);
   } while (svptest_any(svptrue_b64(), pg));
 
@@ -99,7 +100,7 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
 
   int32_t N = n;
   int32_t j = 0;
-  svbool_t pg = svwhilelt_b32(j, N);
+  svbool_t pg = svwhilelt_b32((uint32_t)j, (uint32_t)N);
   int32_t active = svcntp_b32(svptrue_b32(), pg);
   svint32_t index_neg = svindex_s32(0, -1);
   svint32_t index = svindex_s32(0, 1);
@@ -133,7 +134,7 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
     posX += sve_size;
     posX_vec = svdup_s32(posX);
     j += sve_size;
-    pg = svwhilelt_b32(j, N);
+    pg = svwhilelt_b32((uint32_t)j, (uint32_t)N);
     active = svcntp_b32(svptrue_b32(), pg);
   } while (svptest_any(svptrue_b32(), pg));
 

kernel/arm64/trmm_lncopy_sve_v1.c

@@ -52,11 +52,11 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
     FLOAT *ao;
 #ifdef DOUBLE
     svint64_t index = svindex_s64(0LL, lda);
-    svbool_t pn = svwhilelt_b64(js, n);
+    svbool_t pn = svwhilelt_b64((uint64_t)js, (uint64_t)n);
     int n_active = svcntp_b64(svptrue_b64(), pn);
 #else
     svint32_t index = svindex_s32(0, lda);
-    svbool_t pn = svwhilelt_b32(js, n);
+    svbool_t pn = svwhilelt_b32((uint64_t)js, (uint64_t)n);
     int n_active = svcntp_b32(svptrue_b32(), pn);
 #endif
     do
@@ -123,11 +123,11 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
         posY += n_active;
         js += n_active;
 #ifdef DOUBLE
-        pn = svwhilelt_b64(js, n);
+        pn = svwhilelt_b64((uint64_t)js, (uint64_t)n);
         n_active = svcntp_b64(svptrue_b64(), pn);
     } while (svptest_any(svptrue_b64(), pn));
 #else
-        pn = svwhilelt_b32(js, n);
+        pn = svwhilelt_b32((uint64_t)js, (uint64_t)n);
         n_active = svcntp_b32(svptrue_b32(), pn);
     } while (svptest_any(svptrue_b32(), pn));
 #endif

kernel/arm64/trmm_ltcopy_sve_v1.c

@@ -51,10 +51,10 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
     FLOAT *ao;
     js = 0;
 #ifdef DOUBLE
-    svbool_t pn = svwhilelt_b64(js, n);
+    svbool_t pn = svwhilelt_b64((uint64_t)js, (uint64_t)n);
     int n_active = svcntp_b64(svptrue_b64(), pn);
 #else
-    svbool_t pn = svwhilelt_b32(js, n);
+    svbool_t pn = svwhilelt_b32((uint64_t)js, (uint64_t)n);
     int n_active = svcntp_b32(svptrue_b32(), pn);
 #endif
     do
@@ -122,11 +122,11 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
         posY += n_active;
         js += n_active;
 #ifdef DOUBLE
-        pn = svwhilelt_b64(js, n);
+        pn = svwhilelt_b64((uint64_t)js, (uint64_t)n);
         n_active = svcntp_b64(svptrue_b64(), pn);
     } while (svptest_any(svptrue_b64(), pn));
 #else
-        pn = svwhilelt_b32(js, n);
+        pn = svwhilelt_b32((uint64_t)js, (uint64_t)n);
         n_active = svcntp_b32(svptrue_b32(), pn);
     } while (svptest_any(svptrue_b32(), pn));
 #endif

kernel/arm64/trmm_uncopy_sve_v1.c

@@ -52,11 +52,11 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
     FLOAT *ao;
 #ifdef DOUBLE
     svint64_t index = svindex_s64(0LL, lda);
-    svbool_t pn = svwhilelt_b64(js, n);
+    svbool_t pn = svwhilelt_b64((uint64_t)js, (uint64_t)n);
     int n_active = svcntp_b64(svptrue_b64(), pn);
 #else
     svint32_t index = svindex_s32(0, lda);
-    svbool_t pn = svwhilelt_b32(js, n);
+    svbool_t pn = svwhilelt_b32((uint64_t)js, (uint64_t)n);
     int n_active = svcntp_b32(svptrue_b32(), pn);
 #endif
     do
@@ -123,11 +123,11 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLON
         posY += n_active;
         js += n_active;
 #ifdef DOUBLE
-        pn = svwhilelt_b64(js, n);
+        pn = svwhilelt_b64((uint64_t)js, (uint64_t)n);
         n_active = svcntp_b64(svptrue_b64(), pn);
     } while (svptest_any(svptrue_b64(), pn));
 #else
-        pn = svwhilelt_b32(js, n);
+        pn = svwhilelt_b32((uint64_t)js, (uint64_t)n);
         n_active = svcntp_b32(svptrue_b32(), pn);
     } while (svptest_any(svptrue_b32(), pn));
 #endif